Wednesday, June 24, 2026
Home Blog Page 198

Rethinking VM information safety in cloud-native environments

0

VMs outlined by Kubernetes sources

The primary large distinction is in illustration. In conventional virtualization methods, a VM is outlined by an object or set of objects tightly managed by the hypervisor. Its configuration, disk information, snapshots, and runtime state are all saved in a platform-specific means, enabling constant backup semantics throughout totally different environments.

KubeVirt depends on the Kubernetes mannequin as an alternative. Digital machines are outlined utilizing Kubernetes customized sources resembling VirtualMachine, VirtualMachineInstance, and (with CDI) DataVolume, that are saved within the Kubernetes management aircraft. Their configuration is thus described declaratively in YAML, and their life cycle is managed by KubeVirt’s controllers. A VM definition in KubeVirt is subsequently not a bundle of hypervisor objects, however a set of Kubernetes sources describing compute, storage, networking, initialization, and storage volumes.

A technology of Kubernetes directors have come to understand Kubernetes’ open, declarative mannequin and YAML-based definitions, however for VM directors it could be a bit complicated at first. Extra importantly for our functions, the way in which this vital metadata is backed up and restored is totally totally different. You’ll want to make use of Kubernetes-specific instruments quite than the instruments you’ve been utilizing, and people instruments would require not less than a primary understanding of the Kubernetes management aircraft.

10 GitHub Repositories to Grasp OpenClaw



Picture by Writer

 

Introducing OpenClaw

 
OpenClaw is gaining consideration as a framework for constructing autonomous AI brokers that may work together with instruments, run workflows, and automate duties. As an alternative of relying solely on prompts, OpenClaw brokers can execute actions, hook up with exterior providers, and prolong their talents by modular expertise and integrations. Because the ecosystem grows, studying OpenClaw entails understanding extra than simply the core repository.

On this article, we discover 10 GitHub repositories that provide help to grasp OpenClaw. These tasks embody the official repository, guided studying sources, expertise collections, reminiscence programs, and deployment instruments. Collectively, they supply a sensible path for understanding how OpenClaw works and tips on how to construct extra succesful agent programs round it.

 

Mastering OpenClaw with GitHub Repositories

 

// 1. OpenClaw (Official Repository)

The openclaw/openclaw repository is the official place to begin for understanding the OpenClaw challenge. It accommodates the core codebase together with documentation explaining how the agent framework works, the way it connects to exterior fashions, and the way expertise and instruments prolong its capabilities.

Working by the repository helps you perceive the basics of OpenClaw brokers, together with how they execute duties, handle instruments, and work together with exterior providers. The documentation and setup directions present the inspiration wanted earlier than exploring the broader ecosystem of expertise, reminiscence programs, and deployment instruments.

 

// 2. OpenClaw Grasp Expertise

The LeoYeAI/openclaw-master-skills repository focuses on discovering and organizing OpenClaw expertise. Expertise are what flip a fundamental OpenClaw set up into a strong agent able to interacting with exterior instruments, APIs, and providers.

Exploring this repository helps you perceive how the OpenClaw ecosystem extends by modular capabilities. By looking and experimenting with completely different expertise, customers can learn the way brokers work together with instruments and the way actual workflows are constructed across the framework.

 

// 3. Superior OpenClaw Expertise

The VoltAgent/awesome-openclaw-skills repository is among the largest curated collections of OpenClaw expertise. It organizes hundreds of expertise into classes, making it simpler to discover the ecosystem and discover capabilities related to completely different workflows.

This repository is especially helpful for intermediate customers who need to develop their agent’s talents. As an alternative of looking out randomly for instruments, the categorized construction helps you perceive how OpenClaw integrates with exterior programs and the way expertise can rework a easy agent into a flexible automation platform.

 

// 4. Superior OpenClaw Use Circumstances

The hesamsheikh/awesome-openclaw-usecases repository focuses on real-world examples of how OpenClaw brokers are utilized in follow. Moderately than itemizing expertise alone, it highlights sensible workflows and functions that present how the know-how matches into on a regular basis duties.

Learning these examples helps readers transfer from idea to software. It demonstrates how OpenClaw can automate workflows, work together with providers, and help with actual duties, which makes it simpler to know the worth of agent-based programs past experimentation.

 

// 5. Study OpenClaw

The carlvellotti/learn-openclaw repository gives a guided studying path for individuals who need a structured strategy to begin utilizing OpenClaw. As an alternative of exploring the core repo alone, this useful resource focuses on explaining setup, workflows, and sensible utilization patterns in a extra approachable approach.

It helps freshmen transfer from set up to actual utilization by strolling by typical workflows and explaining how OpenClaw matches into on a regular basis automation or assistant duties. For readers preferring tutorials over studying supply code, this sort of guided useful resource makes the educational curve a lot smoother.

 

// 6. memU

The NevaMind-AI/memU repository introduces the idea of persistent reminiscence for AI brokers. It’s designed as a reminiscence layer that enables long-running brokers like OpenClaw to retain context over time as an alternative of relying solely on brief prompts.

Working with reminiscence programs like memU helps readers perceive how brokers can evolve from easy activity executors into proactive assistants. It additionally introduces concepts comparable to long-term context storage, decreased token utilization, and steady agent habits.

 

// 7. ClawRouter

The BlockRunAI/ClawRouter repository focuses on mannequin routing for OpenClaw-style brokers. Routing programs assist decide which AI mannequin ought to deal with a given activity, which might enhance efficiency, value effectivity, and suppleness.

Studying about routing infrastructure helps customers perceive how extra superior agent programs are constructed. As an alternative of counting on a single mannequin, routing permits OpenClaw setups to dynamically choose completely different fashions relying on the duty, making agent architectures extra scalable.

 

// 8. 1Panel

The 1Panel-dev/1Panel repository gives a server management panel designed to simplify self-hosted infrastructure administration. Whereas it isn’t particular to OpenClaw, many customers depend on instruments like 1Panel to deploy and handle providers on digital non-public server (VPS) environments.

Utilizing platforms like 1Panel helps readers learn the way OpenClaw brokers might be hosted and managed reliably. It introduces sensible deployment matters comparable to server administration, container orchestration, and sustaining a steady internet hosting atmosphere for AI instruments.

 

// 9. Umbrel

The getumbrel/umbrel repository is a house server working system designed to run self-hosted functions by a easy app ecosystem. It permits customers to deploy providers from an app store-like interface whereas sustaining full management over their infrastructure.

Exploring Umbrel helps readers perceive how OpenClaw can match right into a broader private server stack. As an alternative of working a single software, customers can construct a whole self-hosted atmosphere the place AI assistants function alongside different providers.

 

// 10. ZeroClaw

The zeroclaw-labs/zeroclaw repository represents the following technology of assistant infrastructure constructed across the OpenClaw ecosystem. The challenge focuses on creating quicker, extra moveable, and extra autonomous assistant programs.

Learning tasks like ZeroClaw helps readers perceive how the ecosystem is evolving. It exhibits how new instruments are pushing agent frameworks towards extra versatile deployment fashions and extra superior automation capabilities.

 

Reviewing the Repositories

 
This desk summarizes what every repository teaches and who it’s best suited to as you discover the OpenClaw ecosystem.

 

Repository What You’ll Study Finest For
openclaw/openclaw Core structure, agent workflows, and the inspiration of the OpenClaw challenge Anybody beginning with OpenClaw
LeoYeAI/openclaw-master-skills Discovering and experimenting with OpenClaw expertise Customers increasing agent capabilities
VoltAgent/awesome-openclaw-skills Giant categorized listing of OpenClaw expertise Intermediate customers exploring the ecosystem
hesamsheikh/awesome-openclaw-usecases Actual-world workflows and sensible functions Customers in search of inspiration for automation
carlvellotti/learn-openclaw Guided studying path and sensible setup directions Novices studying OpenClaw
NevaMind-AI/memU Persistent reminiscence programs for long-running AI brokers Builders constructing proactive brokers
BlockRunAI/ClawRouter Mannequin routing and superior agent infrastructure Superior OpenClaw setups
1Panel-dev/1Panel VPS deployment and server administration for self-hosted instruments Customers internet hosting OpenClaw on servers
getumbrel/umbrel Constructing a broader self-hosted private server stack Customers creating full house server setups
zeroclaw-labs/zeroclaw Rising assistant infrastructure and future ecosystem instruments Readers exploring the place the ecosystem is heading

 
 

Abid Ali Awan (@1abidaliawan) is a licensed knowledge scientist skilled who loves constructing machine studying fashions. Presently, he’s specializing in content material creation and writing technical blogs on machine studying and knowledge science applied sciences. Abid holds a Grasp’s diploma in know-how administration and a bachelor’s diploma in telecommunication engineering. His imaginative and prescient is to construct an AI product utilizing a graph neural community for college kids fighting psychological sickness.

Photo voltaic cells simply did the “unimaginable” with this 130% breakthrough

0


Solar energy performs a significant function in efforts to cut back dependence on fossil fuels and tackle local weather change. The Solar delivers an immense quantity of power to Earth each second, but fashionable photo voltaic cells seize solely a small share of it. This limitation is because of a long-standing “bodily ceiling” that has been tough to beat.

In analysis revealed within the Journal of the American Chemical Society on March 25, scientists from Kyushu College in Japan, working with collaborators at Johannes Gutenberg College (JGU) Mainz in Germany, developed a brand new strategy to push previous this barrier. They used a molybdenum-based metallic advanced generally known as a “spin-flip” emitter to seize further power generated via singlet fission (SF), usually described as a “dream expertise” for bettering mild conversion.

With this strategy, the staff achieved power conversion efficiencies of round 130%, exceeding the standard 100% restrict and pointing towards extra superior photo voltaic applied sciences.

How Photo voltaic Cells Work and Why Power Is Misplaced

Photo voltaic cells produce electrical energy when photons from daylight hit a semiconductor and switch power to electrons, setting them in movement and creating an electrical present. This course of might be in comparison with a relay, the place power is handed from one particle to a different.

Nevertheless, not all photons are equally helpful. Low-energy infrared photons would not have sufficient power to activate electrons, whereas high-energy photons similar to blue mild lose their further power as warmth. Due to this, photo voltaic cells can solely make the most of about one-third of incoming daylight. This constraint is named the Shockley-Queisser restrict and has remained a significant problem.

Singlet Fission Provides a Method To Multiply Power

“We’ve got two major methods to interrupt via this restrict,” says Yoichi Sasaki, Affiliate Professor at Kyushu College’s School of Engineering. “One is to transform lower-energy infrared photons into increased power seen photons. The opposite, what we discover right here, is to make use of SF to generate two excitons from a single exciton photon.”

Below regular circumstances, every photon produces just one spin-singlet exciton after excitation. With SF, this single exciton can cut up into two lower-energy spin-triplet excitons, which may successfully double the accessible power. Though sure supplies similar to tetracene can assist this course of, capturing these excitons effectively has confirmed tough.

Overcoming Power Loss From FRET

“The power might be simply ‘stolen’ by a mechanism referred to as Förster resonance power switch (FRET) earlier than multiplication happens,” Sasaki explains. “We due to this fact wanted an power acceptor that selectively captures the multiplied triplet excitons after fission.”

To handle this problem, the researchers turned to metallic complexes, which might be exactly engineered. They recognized a molybdenum-based “spin-flip” emitter as an efficient answer. On this system, an electron modifications its spin throughout absorption or emission of near-infrared mild, permitting it to seize the triplet power generated by SF.

By fastidiously adjusting the power ranges, the staff minimized losses from FRET and enabled environment friendly extraction of the multiplied excitons.

Collaboration and Experimental Success

“We couldn’t have reached this level with out the Heinze group from JGU Mainz,” Sasaki says. Adrian Sauer, a graduate scholar from the group visiting Kyushu College on change and the paper’s second creator, introduced the staff’s consideration to a fabric lengthy studied there, resulting in the collaboration.

When mixed with tetracene-based supplies in answer, the system efficiently harvested power with quantum yields of about 130%. Because of this roughly 1.3 molybdenum-based metallic complexes have been activated for each photon absorbed, exceeding the same old restrict and demonstrating that extra power carriers have been produced than incoming photons.

Future Photo voltaic and Quantum Know-how Purposes

This analysis introduces a brand new technique for amplifying excitons, though it’s nonetheless on the proof-of-concept stage. The staff goals to combine these supplies into solid-state techniques to enhance power switch and transfer nearer to sensible photo voltaic cell purposes.

The findings may additionally encourage additional analysis combining singlet fission and metallic complexes, with potential makes use of not solely in photo voltaic power but additionally in LEDs and rising quantum applied sciences.

AI fuels a brand new wave of technical debt

0


Fragile techniques, inefficient workflows and strategic gridlock are only a few of the disagreeable unintended effects ensuing from technical debt. These issues can undermine efficiency and undercut innovation. However as CIOs try and navigate this more and more difficult area, they encounter a brand new foe: AI.

What makes AI so difficult is that it behaves in a different way from different digital applied sciences — and it could actually function an accelerant to debt. Legacy techniques, siloed information, outmoded APIs and outdated architectures create a debt basis. AI exposes and amplifies these points, whereas introducing a brand new tax that stretches throughout an enterprise — and right into a provide chain.

“AI funding is not simply one other IT funding; it’s a reinvention of how the enterprise operates,” mentioned Matt Lyteson, CIO of expertise platform transition at IBM. A 2025 research performed by the IBM Institute for Enterprise Worth discovered that of the 1,300 senior AI decision-makers surveyed, those that reported their corporations ignored the difficulty of technical debt noticed returns on initiatives drop by 18% to 29%, with timelines increasing by as a lot as 22%, In the meantime, a Forrester report discovered that 75% of expertise decision-makers count on technical debt to rise to a “extreme” stage in 2026.

Associated:The sunsetting of Sora: A tough lesson in AI portfolio resilience

CIOs could also be on the hook for AI debt, however the issue — and the answer — extends past IT. “There are two elements of the equation,” mentioned Koenraad Schelfaut, a senior managing director at Accenture. “The primary is your present technical debt, which is stopping you from deploying AI at scale. The second is that whereas deploying AI, issues that weren’t technical debt change into technical debt.”

On the margins

At first look, AI-specific debt resembles different sorts of technical debt. It slows groups down, inflates budgets and short-circuits transformation. However AI dials up the challenges: growing older code, undocumented techniques and siloed information broaden from an IT headache to a full-blown enterprise downside. As a result of AI reshapes workflows throughout items and departments, CIOs should look at it by a broader lens of change administration and alternative prices.

The results of this debt compound shortly. “It is not clear who owns, pays and helps AI initiatives,” mentioned Carlos Casanova, a principal analyst at Forrester. This makes it tough to pin down the supply of an issue — or establish the appropriate final result. What’s extra, not like an on-premises server or infrastructure within the cloud, AI debt is usually invisible — till a challenge goes astray, a safety hole seems or a finances overrun surfaces.

Associated:Gartner delivers CIO information to deploying rising expertise

AI debt typically hides behind early success, Schelfaut mentioned. Chatbots help employees, pilot initiatives present promise and preliminary rollouts ship progress. Initiatives achieve momentum, and enterprise leaders achieve confidence. Then, out of the blue, because the group makes an attempt to scale an initiative, issues go astray. “All of the sudden, you’ll be able to’t get techniques to speak to at least one one other, and you’ll’t accomplish what you had got down to do,” he mentioned.

A part of the issue is how CIOs body the difficulty. Many view AI debt as an IT upkeep downside quite than a enterprise problem, Schelfaut mentioned. Consequently, they deal with the price of sustaining legacy techniques however overlook the obstacles they impose. AI flips this logic. “Technical debt is much less about what outdated techniques are costing you to keep up than what they are not permitting you to do,” he mentioned.

Escaping this myopia begins with an understanding of what technical debt really prices, Schelfaut mentioned. He recognized the next 4 distinct dimensions:

  • The direct price of operating and sustaining techniques and infrastructure.

  • The curiosity price related to inefficiencies that stretch over time.

  • Legal responsibility prices associated to safety, compliance and resilience dangers.

  • The chance prices that make it not possible for a corporation to construct out AI.

Most organizations deal with solely the primary dimension, Schelfaut mentioned. The opposite three are the place AI debt does the true harm.

New guidelines, new instruments

Issues aren’t going to get any simpler within the months and years forward. In accordance with the IBM Institute for Enterprise Worth survey, 69% of executives consider that unaddressed technical debt will render some AI initiatives financially untenable. “CIOs and CFOs have to be speaking about debt-adjusted ROI now,” Lyteson mentioned. 

Agentic AI raises the stakes as a result of it introduces new dangers — and publicity factors. Permissions and controls designed for people typically break down when brokers function at machine velocity. And since these brokers talk with one another in methods which might be tough to foretell and monitor, compute and token prices can spiral, driving the necessity for AgentOps alongside FinOps.

As brokers proliferate, conventional monitoring instruments fall quick. New metrics and monitoring instruments should ship visibility into AI agent conduct, interactions and the infrastructure, information and fashions they eat. With out this visibility, CIOs cannot clarify prices, dangers or failures to the board, Casanova mentioned. Additionally they cannot intervene earlier than points set off compliance, safety or operational failures. 

The repair is not extra expertise; it is higher visibility into AI and the workflows it touches. Lyteson mentioned an important start line is to reexamine the way in which initiatives unfold — and who’s answerable for them. IBM makes use of “AI fusion groups” that span IT and enterprise capabilities. These teams “outline the outcomes we need to obtain by AI, run fast experiments to gauge how they affect workflow and have interaction staff to see precisely how their work adjustments,” he mentioned.

As IBM spins up AI initiatives, it measures their worth towards three standards — utilizing every as a software to identify technical debt. Productiveness instruments should display time financial savings. Agentic workflows are held to a unique customary: measurable beneficial properties in income development, operational effectivity or per-unit workflow prices. Compliance and safety initiatives should present a transparent discount in danger.

Balancing the books

The thought is not to eradicate technical debt earlier than deploying AI, Schelfaut mentioned. It is to establish obstacles to progress and engineer important fixes. This requires abandoning the mindset that new AI options can sit straight atop present infrastructure and performance inside point-to-point interfaces. The excellent news? AI itself is an efficient software for figuring out points — documenting legacy techniques, rewriting fragile code and figuring out what structure wants to vary.

A robust governance framework is the glue that holds all the pieces collectively, Casanova mentioned. As AI instruments multiply throughout IT and enterprise items, organizations should absolutely perceive hidden infrastructure prices, information sovereignty, entry permissions and controls, AI sprawl and IP leakage. “If somebody creates an agent, maybe it ought to go right into a repository for vetting earlier than it is deployed,” he mentioned.

Ultimately, CIOs should acknowledge that AI technical debt is not an issue to resolve — it is a situation to handle. Throwing expertise on the problem will not pay down the debt. “It is about greater than transformation,” Lyteson concluded. “It’s about steady enchancment. You want a framework that’s adequate to start out and versatile sufficient to refine, so you’ll be able to iterate on what’s working and weed out what shouldn’t be.”



A girl’s uterus has been stored alive exterior the physique for the primary time


“As a proof of idea, it’s spectacular,” says Keren Ladin, a bioethicist who has targeted on organ transplantation and perfusion at Tufts College. “These are early days.”

It won’t sound like a lot, however 24 hours is a very long time for an organ to be out of the physique. Sustaining a donated uterus for that lengthy may develop the choices for uterus transplant, a reasonably new process supplied to some individuals who wish to be pregnant however don’t have a useful uterus, says Gerald Brandacher, professor of experimental and translational transplant surgical procedure on the Medical College of Innsbruck in Austria.

“It’s higher than what we at the moment have, as a result of we have now solely a few hours,” he says. Up to now, most uterus transplants have been deliberate operations involving organs from residing donors. A expertise like this might enable for the usage of extra organs from deceased donors, he says.

That work is “not within the rapid pipeline” for the staff in Spain, says Santamaria. “We’re engaged on different issues.”

Being pregnant within the lab?

Santamaria, González, and their colleagues are extra serious about utilizing sustained human uteruses for analysis. 

They’ve mounted a digicam to a wall within the nook of the room, pointed at their machine. It permits the staff to watch “Mom” remotely, and to examine if any valves disconnect. (That occurred as soon as earlier than—a spike in stress brought about the blood bag to come back unfastened, spilling a liter of blood on the ground, Santamaria says.)

They’d like to have the ability to maintain their uteruses alive for round 28 days to check the menstrual cycle and problems that have an effect on the uterus, like endometriosis and fibroids.

New Infinity Stealer malware grabs macOS knowledge by way of ClickFix lures

0


A brand new info-stealing malware named Infinity Stealer is focusing on macOS techniques with a Python payload packaged as an executable utilizing the open-source Nuitka compiler.

The assault makes use of the ClickFix method, presenting a pretend CAPTCHA that mimics Cloudflare’s human verification verify to trick customers into executing malicious code.

Researchers at Malwarebytes say that is the primary documented macOS marketing campaign combining ClickFix supply with a Python-based infostealer compiled utilizing Nuitka.

As a result of Nuitka produces a local binary by compiling the Python script into C code, the ensuing executable is extra proof against static evaluation.

In comparison with PyInstaller, which bundles Python with bytecode, it’s extra evasive as a result of it produces an actual native binary with no apparent bytecode layer, making reverse engineering a lot more durable.

“The ultimate payload is written in Python and compiled with Nuitka, producing a local macOS binary. That makes it more durable to research and detect than typical Python-based malware,” Malwarebystes says.

Assault chain

The assault begins with a ClickFix lure on the area update-check[.]com, posing as a human verification step from Cloudflare and asking the consumer to finish the problem by pasting a base64-obfuscated curl command into the macOS Terminal, bypassing OS-level defenses.

The ClickFix step
ClickFix step utilized in Infinity assaults
Supply: Malwarebytes

The command decodes a Bash script that writes the stage-2 (Nuitka loader) to /tmp, then removes the quarantine flag, and executes it by way of ‘nohup.’ Lastly, it passes the command-and-control (C2) and token by way of surroundings variables after which deletes itself and closes the Terminal window.

The Nuitka loader is an 8.6 MB Mach-O binary that incorporates a 35MB zstd-compressed archive, containing the stage-3 (UpdateHelper.bin), which is the Infinity Stealer malware.

The malware's disassembly view
The malware’s disassembly view
Supply: Malwarebytes

Earlier than beginning to accumulate delicate knowledge, the malware performs anti-analysis checks to find out whether or not it’s working in a virtualized/sandboxed surroundings.

Malwarebytes’ evaluation of the Python 3.11 payload uncovered that the info-stealer can take screenshots and harvest the next knowledge:

  • Credentials from Chromium‑based mostly browsers and Firefox
  • macOS Keychain entries
  • Cryptocurrency wallets
  • Plaintext secrets and techniques in developer recordsdata, comparable to .env

All stolen knowledge is exfiltrated by way of HTTP POST requests to the C2, and a Telegram notification is distributed to the menace actors upon completion of the operation.

Malwarebytes underlines that the looks of malware like Infinity Stealer is proof that threats to macOS customers are solely getting extra superior and focused.

Customers ought to by no means paste into Terminal instructions they discover on-line and don’t absolutely perceive.

Automated pentesting proves the trail exists. BAS proves whether or not your controls cease it. Most groups run one with out the opposite.

This whitepaper maps six validation surfaces, reveals the place protection ends, and offers practitioners with three diagnostic questions for any instrument analysis.

A intestine microbe linked to the Mediterranean eating regimen boosts muscle energy in mice

0

Individuals with stronger muscle tissues usually tend to harbor a selected species of micro organism of their guts, and when this bacterial species was fed to mice, they grew to become stronger, a brand new research finds.

The research authors say the microbe has the potential to be a part of a probiotic complement, probably boosting muscle energy. Nonetheless, this might require the researchers to discover a technique to protect it in a tablet. What’s extra, this microbe might function a drug to deal with frailty within the aged, assuming future scientific trials in people reveal the microbe safely improves muscle energy, mentioned research lead writer Borja Martinez-Tellez, a sports activities scientist at Leiden College within the Netherlands.

Computing sine and cosine of advanced arguments with solely actual features

0




Suppose you might have a calculator or math library that solely handles actual arguments however you want to consider sin(3 + 4i). What do you do?

Should you’re utilizing Python, for instance, and also you don’t have NumPy put in, you need to use the built-in math library, nevertheless it won’t settle for advanced inputs.

>>> import math
>>> math.sin(3 + 4j)
Traceback (most up-to-date name final):
File "", line 1, in 
TypeError: should be actual quantity, not advanced

You need to use the next identities to calculate sine and cosine for advanced arguments utilizing solely actual features.

The proof could be very easy: simply use the addition formulation for sine and cosine, and the next identities.

begin{align*} sin iz &= i sinh z  cos iz &= cosh z end{align*}

The next code implements sine and cosine for advanced arguments utilizing solely the built-in Python features that settle for actual arguments. It then exams these in opposition to the NumPy variations that settle for advanced arguments.


from math import *
import numpy as np

def complex_sin(z):
    x, y = z.actual, z.imag
    return sin(x)*cosh(y) + 1j*cos(x)*sinh(y)

def complex_cos(z):
    x, y = z.actual, z.imag
    return cos(x)*cosh(y) - 1j*sin(x)*sinh(y)

z = 3 + 4j
mysin = complex_sin(z)
mycos = complex_cos(z)
npsin = np.sin(z)
npcos = np.cos(z)
assert(abs(mysin - npsin) < 1e-14)
assert(abs(mycos - npcos) < 1e-14)

Associated posts





Saturday morning’s tabs!

0


It’s thay day of the week the place I take a tab, I take a look at it, I say a prayer, a lament, after which put up it right here simply as I delete it, by no means to be seen once more.

Scott’s Mixtape Substack is a reader-supported publication. To obtain new posts and help my work, think about turning into a free or paid subscriber.


The Celtics beat the Hawks. Pritchard dropped like one million factors. It was a enjoyable sport. I considered Rodney Andrews and laughed pondering of all of the nasty issues he would’ve mentioned watching the Hawks not pull by.

Wage freezes and hiring cuts at Princeton.

Claude Code’s auto mode could also be a greater match for many who don’t need to reside so dangerously.

Extra about chatbot psychosis, or the formation of it, from a comfort pattern and no less than not a pure anecdote. This Stanford staff analyzed chat logs from 19 customers who skilled psychological hurt from LLM chatbots, discovering that chatbots have been sycophantic in over 80% of messages and persistently misrepresented their very own sentience or capabilities. These have been patterns that correlated with customers creating delusional pondering, romantic attachments, and for much longer conversations. The researchers additionally discovered that when customers expressed violent ideas, chatbots inspired or facilitated these ideas in roughly a 3rd of circumstances , elevating severe considerations about present safeguards.​​​​​​​​​​​​​​​​

Meek Mill has discovered Claude and it’s altering his life.

Dr. Jim O’Connell was a Harvard-trained doctor who in 1985 primarily gave up a prestigious fellowship to discovered Boston Well being Take care of the Homeless and by no means regarded again. Following him was a brand new affected person named Tony, contemporary out of jail and by some means wired into each social community on Boston’s streets, walks into the Thursday Road Clinic trailing sweat, charisma, and feathers from a torn parka. What will get me is the ending element Kidder leaves you with: after Tony leaves, all that is still of him within the examination room are a couple of feathers of down from the holes in his parka, white fluff on the pale inexperienced linoleum ground.

Andrew McCarthy felt disconnected from his associates as he aged, so he made extra intentional efforts to spend most time with them on many adventures. He has a brand new guide known as Who Wants Buddies and I purchased it yesterday.

A new replace to your Apple working system features a graphic on the high that may inform you about whether or not your battery can’t get charging.

A brand new Apple Imaginative and prescient Professional replace brings new visuals that might type the basis for even richer, immersive experiences. I really like they’ve refused to surrender this phenomenal machine. It’s greatest Apple product they’ve had for the reason that iPhone.

Lengthy essay that Elon has made all the correct massive bets and that xAI will dominate.

J Cole calls the mass cancelation by Kendrick and others who piled on of Drake disgusting. Put it to J Cole — he follows the beat of his personal drummer.

Apparently, Hinge has an aggressive nearly “one strike and also you’re out” coverage and generally an individual can’t determine what they did, and the corporate doesn’t have a system to resolve it simply. It’s attention-grabbing to think about that Match is mainly a oligopolist. They personal all the most important relationship apps aside from Bumble, which got here out of Tinder and was the one who obtained away. So when you get banned from one, you get banned from almost all relationship platforms, which is the dominant platform for assembly and discovering romance. That’s quite a lot of energy one firm can have over someone’s life.

CodeChella Madrid (third annual) is formally offered out! With two months to go making this our hottest venue but. Please come subsequent yr; plus we can have a quickly to be introduced second CodeChella someplace else. So keep tuned.

A shiny app on hashish use and diff-in-diff.

Anthropic’s interpretability analysis hub is a part of their ongoing try and crack open the black field of huge language fashions and determine what’s really occurring inside them computationally. They’ve launched attention-grabbing outcomes from this lab for a couple of years. The premise they lead with is sort of disarmingly trustworthy: a shocking truth about fashionable LLMs is that no one actually is aware of how they work internally , and this staff is making an attempt to vary that, paper by paper, going again to 2021. The current work is genuinely fascinating. In a number of I’ve seen, they’ve printed findings suggesting Claude can introspect by itself inside states, traced how consideration patterns emerge from characteristic interactions, and even mapped the geometry underlying one thing as basic as how a mannequin counts.

Do you surprise how an airfoil causes flight? Do you need to see a cool graphic? Good as a result of that is one.

Is economics software program engineering discipline? Possibly. However with Claude code doing all software program engineering, what then does that make economics going ahead?

Most Progress
Economics is a Discipline of Software program Engineering
What does it imply to do empirical social science…
Learn extra

Consideration (and cash) is all you want. Nice title for a brand new NBER working paper on the way it’s not possible for universities to rent the brand new AI expertise. Extra information to place in my head that the unraveling of human historic establishments of science are persevering with to be stretched aside.

One actor makes use of social media and blocks folks. Sure, this was a narrative.

A federal choose blocks DoD’s retaliation in opposition to Anthropic. For now.

Figuring out prediction errors in observational knowledge by Ashesh Rambachan, one of the proficient, humble younger econometricians doing sensible and beneficial work for the utilized group.

Ben Affleck low constructed a 16-person AI firm and quietly offered it to Netflix for $600 million. WOW.

Are two thirds of Gen X ladies going through psychological well being issues? Must dig into that.

Previous Michael Jordan tales from Vince Carter.

Is it the loss of life of the romantic comedy style? I wouldn’t rely that one out, however right here’s an Atlantic piece about its demise.

Don’t neglect — award profitable texas bbq joint, Helberg, situated in Waco Texas delivers wherever within the nation. Think about what they’ll say once you present as much as your Boston marathon get together with it!

Does AI want a structure? Possibly says the New Yorker.

I proceed to maintain my eye on this queen sized mattress body.

Don’t neglect, Schmidt Ventures is giving grants out to review AI and work. Don’t let the deadline get away from you.

Hacking and phishing has seen a better yield on effort after Gen AI. Many ways now rising we’ve by no means seen earlier than. Bear in mind, our mother and father are weak resulting from cognitive decline. Behavioral analytics is due to this fact an necessary a part of defending ourselves from it, so when you have a background in that, think about this as doable profession route.

Are the brokers.md markdowns even serving to us in any respect with coping with the AI brokers?

Saturn Devouring His Son.

My penalized regression slides from this weeks undergrad class — day one and day two — turned out good I believed. See my dialogue of andrew bakers work in direction of the top of this speak. I discovered quite a bit instructing this class, in addition to my PhD likelihood class, for which I’ll be perpetually grateful.

I actually loved engaged on yesterday’s put up concerning the papers at Zurich’s APE undertaking of AI automated Econ manuscripts. I discovered large proof of p-hacking.

Andy Corridor at Stanford can also be taking a look at this, and has not discovered p-hacking in his bots, however I spoke with him yesterday, and it’s clearer to me his method is completely different. He’s giving them the dataset and the estimation. Whereas APE is totally automated, the concept, discovering the information, creating the identification technique and the estimator. And in Nick Huntington-Klein’s work, on the numerous analyst design, he has additionally discovered that the researcher levels of freedom occur far additional up the pipeline than estimation. It’s moderately within the cleansing and preparation phases. So these should not contradictory findings about AI brokers p-hacking. It’s nonetheless attention-grabbing although, and I needed to say thanks to David Yanagizawa-Drott, who runs the Social Catalyst Lab, the APE undertaking, and was the Yrjö Jahnsson Award winner for 2025, for giving me the inexperienced gentle to look into what they’re doing. David’s undertaking is leading edge and engaging. And I extremely encourage you to examine him out.

I gave a chat yesterday for the Board of Governors on AI brokers that was acquired properly. I known as it “AI Brokers for Analysis Employees”, an homage to Ronald Fisher’s 1925 guide, “Statistical Strategies for Analysis Employees”. I believe I’m leaning in direction of a guide of the identical title. I’ll be presenting variations of the speak, and including to it, till I see if it seems like there’s a desk of contents buried in all my ideas and writings and talks about it. So want me luck!

However within the meantime, the Remix comes out this summer time. I obtained the proofs and ship them again final week. What a journey. So glad it’s performed. It’s my love letter to the Princeton Industrial Relations Part, the Harvard stats and Econ dept, all from the Nineteen Seventies by the Nineties, and it’s about as updated and intermediate stage in nature as I may make it. It’s clocked in at 750 pages which in fact means it’s filled with my typical rambling.

I’m right down to 4 weeks earlier than my time at Harvard is up. I’m considering holding my house right here on Comm Ave and commuting. I don’t assume I’m fairly able to let go and the chums I’ve made. So we’ll see.

And with that, most of my hyperlinks are gone. Not less than those on this telephone. Hope everybody has an ideal weekend. Spring is upon us!

Scott’s Mixtape Substack is a reader-supported publication. To obtain new posts and help my work, think about turning into a free or paid subscriber.

Athena: Intermediate Representations for Iterative Scaffolded App Technology with an LLM

0


It’s difficult to generate the code for a whole consumer interface utilizing a Giant Language Mannequin (LLM). Person interfaces are complicated and their implementations typically encompass a number of, inter-related recordsdata that collectively specify the contents of every display, the navigation flows between the screens, and the information mannequin used all through the appliance. It’s difficult to craft a single immediate for an LLM that comprises sufficient element to generate a whole consumer interface, and even then the result’s continuously a single giant and obscure file that comprises the entire generated screens. On this paper, we introduce Athena, a prototype software era setting that demonstrates how using shared intermediate representations, together with an app storyboard, information mannequin, and GUI skeletons, might help a developer work with an LLM in an iterative style to craft a whole consumer interface. These intermediate representations additionally scaffold the LLM’s code era course of, producing organized and structured code in a number of recordsdata whereas limiting errors. We evaluated Athena with a consumer examine that discovered 75% of members most popular our prototype over a typical chatbot-style baseline for prototyping apps.