X is proud to be partnering with @au_official as an essential app for users in Japan. Starting today au Starlink Direct customers can stay connected to X in mobile dead zones. This data service is powered by @Starlink Direct to Cell satellites. https://t.co/hbdHEyCtCa

UPDATE: https://t.co/vqCVcm0EQl

🚀Supercharge your developer workflow with local and remote MCP Servers inside VS Code and GitHub Copilot Agent mode! ▶️https://t.co/AlWmBHMzM9 #vscode #mcpserver #githubcopilot https://t.co/5p5HiCwh1e

📢 VS Code is live in 10 minutes! @pappapez is here to demo how to use #joyride and #copilot to script VS Code in user space,! Join us and chat live with the team 🎬 https://t.co/GEq0L93CQr https://t.co/LIeKdknzt0

I know I'm sounding like a fangirl, which I suppose I am, whenever it comes to @code and @GitHubCopilot, but I really think people are missing out. One aspect that I think is worth paying attention to is how they structure their subscriptions. IMO this is not only very good deal, but the most transparent and honest subscription in the market. Consumption is measures in requests (not tokens), with clear explanation of how requests are counted — a single user interaction in the chat, a single run of the remote coding agent (which is amazing, it can get a lot done in one go), and a multiplier for "premium" models (for example: 1x for GPT-5, Claude Sonnet or Gemini 2.5 Pro, 10x for Claude Opus, 0.25x for Grok). Copilot provides access to the best models from all providers, as well as the ability to connect to any API or local model with your own key. The affordable base subscriptions include unlimited usage of smaller models like GPT-4.1 and GPT-5-mini (both really powerful and great for lots of tasks) and a base allowance of premium requests. Additional requests are metered per consumption and reported clearly in the UI. The result is a package that's both a great deal and very predictable and transparent. I really think they set the bar for how AI subscriptions should be structured. See https://t.co/L7NMLc1ZW7 and https://t.co/KX68MUW7uK * full disclosure: I have been gifted a pro subscription ($10) as an active open-source contributor and, more recently, a free upgrade to pro+ ($39) so that I can try advanced features. I would have gladly paid for these out of pocket, as I'm certain I will when my freebies run out, and I am paying for extras when I hit the limits.

+2 more

🌑🦉 Calling all you night owls and Playwright fans. Join @debs_obrien and @skn0tt at 1:00 am PT pm Friday August 29th for a demo of #Playwright in #VSCode with the Playwright VS Code Extension! Join us to chat live with the team 🎭 https://t.co/EIHXRVfON3 https://t.co/sZ99gCFa8o

We're so excited to bring you VS Code Dev Days - a global community initiative to explore the AI capabilities within @code! With more than 60 worldwide, in-person events and 4 special virtual editions, there's an event for you. Learn more: https://t.co/bvaTPJ4rDO https://t.co/UQQxKZYpGj

@ggerganov I used to have 2x 4090s on the pc, which definitely did cause a lot of issues - when I tested without the 2nd 4090 back then it sped everything up dramatically. But now, just a single 5090 on here - here's my fire hazard dusty ass rig xD apologies for all the sadness this image will cause people 😂

@ostensiblyneil @ggerganov Isnt that what i posted here Or is it not using it you mean? https://t.co/uwDnt1Ih85

@ostensiblyneil @ggerganov Its this one https://t.co/UaMVD1LlHM

Nous Research introduced Hermes 4, a family of open, hybrid reasoning models Designed to be neutral, the models achieve SOTA on RefusalBench, which tests for willingness to be helpful in scenarios usually disallowed by both closed and open models https://t.co/QU0sMGwWjU https://t.co/Hn7R9071ly

This is also a very interesting test I give image models. Most cannot generate a blank white (or any color) image because it would allow detection of their hidden watermarks. Gemini is no exception, it absolutely refuses to just generate a blank image. https://t.co/QNIgnIOf46

AI godfather Geoffrey Hinton says superintelligence is coming soon, but we have no idea how to control it. "I always thought there'd be plenty of time to figure out what to do about the risks, and there isn't plenty of time anymore." https://t.co/2L0HGyQoCz

The earth’s core is hot. So hot, that if we drilled deep enough, we could power the world millions of times over with cheap, clean energy. https://t.co/RUExESv5T2

One of the best LW posts I've read recently, and an absolute smash cut of an opener https://t.co/oPuplvsFPU https://t.co/rqNJYImLbF

There is a great retrospective in the WSJ today on how the UK's post-Brexit experiment with big increases in legal immigration went wrong. Their mistakes yield some important policy lessons for supporters *and* critics of skilled immigration. Let's dive in: (tl;dr: What was intended to be a skilled immigration surge made the catastrophic mistake of setting visa criteria based on college degrees and workers filling so-called "labor shortages," ultimately leading instead to a surge in low-skilled, rather than high-skilled, immigration.) Post-pandemic Britain dramatically increased net migration, shifting away from the EU and towards migrants from outside Europe. Visa issuance went up across the board: work visas, study visas, and particularly dependent visas. The plan was initially supposed to be a limited, targeted, skilled immigration surge. The UK, post-Brexit, would be free to go out and attract top talent from anywhere. But things went wrong quickly. What constituted "skilled" work included roofers (!) alongside business executives. The system was then blown apart by businesses claiming "labor shortages." Rather than raise pay, businesses appealed for special carve-outs in areas like construction. Basing their visa programs on filling so-called shortages invited this kind of lobbying. Nobody agrees on what a labor shortage actually means. The UK has a technocratic body meant to identify them, but it's ultimately just vibes and raw interest group battles all the way down. Next, the UK rubber-stamped visas for students and their dependents regardless of the quality of the program or school in which they enrolled. There was a subsequent explosion in low-quality degree programs catering to foreign students (and their dependents) to take advantage of this option. From the article: "While some students went to well-known universities such as Oxford or Cambridge, the number of students undertaking one-year master’s degrees from lesser known British universities shot up. And unlike pre-Brexit, many more stayed on in the U.K. after graduating. The fruits of that recruitment drive are visible today. In east London sit two modern high-rises, the Import Building and the Export Building, which house three different universities’ London campuses stacked next to each other." Designing student visa programs this way created an enormous incentive for both colleges and students to expand + take advantage of low-quality degree programs. This is something Tory MP Neil O'Brien covered well in this 2023 Substack post, "The Deliveroo Visa Scandal": https://t.co/G9tICrPEsB The political backlash to all this is growing and, in my view, the UK will end with enormous, long-lasting cuts to immigration in part because of these mistakes (and not building housing, but that's another conversation...). So what lessons does this have for how the US should design skilled immigration policy? Here are a few: 1. Skilled immigration programs need to be ruthlessly and narrowly focused on admitting the highest-paid applicants. High salaries are hard to game or fake. It's the most transparent criteria you can set. 2. Ignore appeals to "labor shortages." This is something that both many proponents and opponents of skilled immigration get wrong. There is no widespread agreement on what an industry or occupation-specific labor shortage actually means. Nor is there much reason to think that filling "shortages" is any better than simply trying to attract the most talented, well-paid people. 3. Do not outsource your immigration system to university admissions offices. Their incentives are not aligned with maximizing the long-term economic and fiscal benefits to the country. "Stapling green cards to diplomas," a longtime zombie proposal from my fellow pro-high-skilled immigration advocates, would create perverse incentives that would be difficult to combat. These are many of the same lessons my colleagues and I wrote about in January in our report, Exceptional By Design, which laid out in detail a new vision for America's high-skilled immigration system. I'm biased, of course, but I think everything we've seen since reinforces what we wrote here. I highly encourage you to check it out: https://t.co/oovumokLfx

+2 more

fantastic work on an important question: which AIs feed user delusions and which tell the user that they're wrong? https://t.co/4ZY645KHzK https://t.co/Y3hzNhItfC

For those who did not read the heatmap study, here is a quick explainer on the wording of the actual question to help you decide where you stand https://t.co/b8NuuW2kqj

Congress has announced an investigation into Wikipedia. The investigation was triggered in large part by my @PirateWires report from October, "How Wikipedia's Pro-Hamas Editors Hijacked the Israel-Palestine Narrative." The piece shows how 40 anti-Israel editors – the Gang of 40 – made a collective 850,000 edits to 10,000 articles, completely reshaping the PIA (Palestine Israel articles) topic area. The Gang of 40 works to severs historical, spiritual and cultural ties between the land and the Jewish people by removing those ties in articles, from the biggest broad-stroke to the most trivial (seeming) minutiae. The main thrust of the Gang of 40, however, is to paint Israel as a "settler colonial" occupier – a goal is has achieved with surprising success. Four months after my piece came out, Wikipedia's "Supreme Court," ArbCom, held one of its most sweeping arbitration proceedings in years. Six leading editors from the G40 were handed indefinite topic ban. One admin was warned by ArbCom. Another editor, who led the Tech For Palestine canvassing effort, was given a site ban – the most severe punishment Wikipedia can deliver. Make no mistake, though, the Gang of 40 is still working hard to achieve its ideological mission. And it's succeeding beyond even their imagination. I'll have more on that very soon. Note: among the many people taking a victory lap on this, you won't see the single most important figure — my source on this piece. I did have more than one source, but one stands out. They know who they are.

Vote for The Courier Journal high school softball Player of the Year for the Louisville area. The poll closes Thursday. Meet the nominees. 📝: @PrinceJStory and @kyhighs / The Courier Journal https://t.co/n5xgv2PafT https://t.co/UNY1jgfydv

Source: https://t.co/D7JGDiwtxT

openai co-founder urges ai labs to conduct safety tests on competing models for responsible business innovation Source: https://t.co/0fwogWLgJP

🧠🧪@haizelabs is automating software testing for AI systems on #AWS. We caught up with Leonard Tang at AWS NYC Summit to learn how the startup helps businesses catch bugs and rigorously test before launch. Find us at AWS Summit LA to connect. 🚀 https://t.co/Q1JERp2Tje https://t.co/jG2z70uwRl

Parallel agents are emerging as an important new direction for scaling up AI. AI capabilities have scaled with more training data, training-time compute, and test-time compute. Having multiple agents run in parallel is growing as a technique to further scale and improve performance. We know from work at Baidu by my former team, and later OpenAI, that AI models’ performance scales predictably with the amount of data and training computation. Performance rises further with test-time compute such as in agentic workflows and in reasoning models that think, reflect, and iterate on an answer. But these methods take longer to produce output. Agents working in parallel offer another path to improve results, without making users wait. Reasoning models generate tokens sequentially and can take a long time to run. Similarly, most agentic workflows are initially implemented in a sequential way. But as LLM prices per token continue to fall — thus making these techniques practical — and product teams want to deliver results to users faster, more and more agentic workflows are being parallelized. Some examples: - Many research agents now fetch multiple web pages and examine their texts in parallel to try to synthesize deeply thoughtful research reports more quickly. - Some agentic coding frameworks allow users to orchestrate many agents working simultaneously on different parts of a code base. Our short course on Claude Code shows how to do this using git worktrees. - A rapidly growing design pattern for agentic workflows is to have a compute-heavy agent work for minutes or longer to accomplish a task, while another agent monitors the first and gives brief updates to the user to keep them informed. From here, it’s a short hop to parallel agents that work in the background while the UI agent keeps users informed and perhaps also routes asynchronous user feedback to the other agents. It is difficult for a human manager to take a complex task (like building a complex software application) and break it down into smaller tasks for human engineers to work on in parallel; scaling to huge numbers of engineers is especially challenging. Similarly, it is also challenging to decompose tasks for parallel agents to carry out. But the falling cost of LLM inference makes it worthwhile to use a lot more tokens, and using them in parallel allows this to be done without significantly increasing the user’s waiting time. I am also encouraged by the growing body of research on parallel agents. For example, I enjoyed reading “CodeMonkeys: Scaling Test-Time Compute for Software Engineering” by Ryan Ehrlich and others, which shows how parallel code generation helps you to explore the solution space. The mixture-of-agents architecture by Junlin Wang is a surprisingly simple way to organize parallel agents: Have multiple LLMs come up with different answers, then have an aggregator LLM combine them into the final output. There remains a lot of research as well as engineering to explore how best to leverage parallel agents, and I believe the number of agents that can work productively in parallel — like the humans who can work productively in parallel — will be very high. [Original text, with links: https://t.co/ElcJZyzcfw ]

AI And IP: Creators Want Protection, But Entrepreneurs Want Speed https://t.co/zm4IKN4AYe @forbes

Study of 67,000 Job Interviews Finds AI Outperforms Human Recruiters https://t.co/WQnROVM8EE @business

More than 10 European startups became unicorns this year https://t.co/VeymmNa4TR @techcrunch @abracarioca

More than 10 European startups became unicorns this year https://t.co/VeymmNa4TR @techcrunch @abracarioca

Revolut Weighs Hiring Adviser for Possible Acquisition in the US https://t.co/SYqgnxn7il @business

Revolut Weighs Hiring Adviser for Possible Acquisition in the US https://t.co/SYqgnxn7il @business

How We Chose the @TIME100 AI 2025 https://t.co/XThlyQlcsE @sampjacobs

ChatGPT isn’t just for cheating anymore https://t.co/euM3QVlqEC @adamclarkestes @voxdotcom