标签：#ai

New frontier model refuses cybersecurity, biology, and chemistry queries.

Grab a refreshing slice of cold watermelon (because it’s summer in Tokyo and I love it!), and let me...

Tuesday's Nintendo Direct showcase felt like an important moment for the company. With the Switch 2 heading into its second holiday season, one in which the hardware will be even more expensive thanks to a price hike, it was a chance for Nintendo to really sell new audiences on its latest console - but that's […]

Voice translations preserve speaker's tone, pacing, pitch—with SynthID watermarks for security.

If those same AI workloads can be handled by cheaper models without affecting quality, it would mean a massive shift in the economics of AI.

Good news for Claude devs deploying on Google Cloud. Claude Fable 5 is now in General Availability (GA) on Google Cloud. You can now access Fable 5 , as well as other Anthropic models - including Claude Opus 4.8 and Claude Sonnet 4.6 - on Agent Platform . Read more here -> [ blog ] Happy building!

Shareable blog post edition: https://andymaleh.blogspot.com/2026/06/andys-laws-of-ai-in-software-engineering.html Law #1: "The more Software Developers use AI, the more valuable Software Engineers who do not use AI become." Software Engineers who are masters at delivering Software without using AI will actually have increased job security the more Software Developers in the worldwide Software Development community rely on AI to deliver Software without having true mastery over Software Engineering. As more Software Developers become fully dependent on AI to build Software without truly understanding how AI gets work done, Software Engineers who do understand what is going on under the hood will dwindle and become more valuable than ever. In other words, they will have a competitive advantage over Software Developers who can only deliver Software features with AI as well as Software Developers who have not mastered Software Engineering. Also, there will always be a need for Software Engineers who can maintain the Software of AI itself. Law #2: "Software Developers benefit from AI in direct proportion to how weak they are in Software Engineering" The weaker Software Developers are at Software Engineering the more they benefit from AI. After all, AI learns from Master Software Engineers and then applies its learnings in code generation done for lower-level Software Developers who lack mastery in Software Engineering. So, users of AI simply place themselves lower in the expertise hierarchy to be on the receiving end of what Master Software Engineers feed AI with their code. This explains why many experts like Linus Torvalds do not find AI very useful while devs who have zero degrees and qualifications feel like they get a lot from AI. A beneficial thing to learn from this law is that it is more valuable for a Software Developer to hone in their Software Engineering skills (including the completion of university degrees) than to hone in their AI usage skills because if t

In Q1 2026, OpenAI and Anthropic moved enterprise customers from flat-rate plans to token-based billing. The change looks administrative, but it had a direct consequence for engineering teams: the real cost of AI became visible for the first time. The market's reaction over the following two months was enough to reopen a question many considered settled: does AI actually deliver measurable ROI? What happened when the bill arrived The most documented case is Uber. The company had encouraged all employees to use agentic tools as much as possible and even ranked AI usage internally on leaderboards. The result: the entire annual budget was consumed in four months. The response was a $1,500/month cap per employee per agentic coding tool (Claude Code, Cursor, and similar). At Brex, engineers were limited to $500/week in tokens; employees outside engineering received a $5/week cap. T-Mobile temporarily capped usage at $2,000/month per user with plans to migrate to a tiered system. One unnamed company, according to Ed Zitron in "AI Is Slowing Down" (June 2026), spent $500 million on Anthropic models in a single month due to absent spend controls. These are not isolated cases. A KPMG survey reported by the Wall Street Journal in June 2026 found that only 26% of companies have a comprehensive view of their AI costs; 50% have partial visibility; and 22% only find out what they owe after the bill arrives. Steve Chase, KPMG's global head of AI, told the Journal: "It's a new resource that needs to be managed that didn't exist quite that way, and we're seeing exponential growth." The structural problem behind the spending caps The spending caps are a symptom. The root cause, as Zitron details in the same article, is that the economics of generative AI require numbers that currently seem out of reach. Anthropics has made over $330 billion in compute commitments with Google, Amazon, and Microsoft, plus another $45 billion with CoreWeave and SpaceX. To cover those commitments, it nee

« Je n'ai fait celle-ci plus longue que parce que je n'ai pas eu le loisir de la faire plus courte. » — Blaise Pascal, Lettres provinciales , Lettre XVI (1656) "I have made this one longer only because I have not had the leisure to make it shorter." Pascal's joke is the whole problem: the short version is the expensive one. LLMs lean the other way, they pad. So the question is whether a model can rein in its own verbosity, and what the trimming costs when the deciding clause is buried: "…shall not disclose, except to affiliates who…" Drop the "except," and the answer flips. The test We use ContractNLI: real NDAs, each with expert Entailment / Contradiction / NotMentioned labels. The clauses that decide a label, the buried "negation-by-exception" conditions, we tag as traps . The metric is decision-survival , and it's judge-free: answer from the full document (the ceiling), compress, answer again, score by exact match against the expert label. Survival is the fraction of full-document-correct answers that stay correct after compression. Compression is blind to the question and computed once per document. Three compressors on Groq ( llama-3.1-8b , qwen3-32b , gpt-oss-120b ), one fixed reader ( llama-3.3-70b ), 400 items across 61 NDAs, two prompts: naive ("Summarise this") and effortful (a careful lossless instruction). The raw ceiling is 66%, but 87% on traps, an artifact of the label mix, which is exactly why we report survival rather than accuracy. Finding 1: Prompt engineering is still alive Decision-survival on trap clauses: Compressor naive effortful llama-3.1-8b 57% 74% qwen3-32b 88% 93% gpt-oss-120b 91% 95% The weak model jumps +16 points on traps; the capable ones improve slightly. The payoff from a better prompt is largest exactly where capacity is scarce. Finding 2: The traps catch out simpler models Decision-survival on ordinary (non-trap) clauses: Compressor naive effortful llama-3.1-8b 87% 87% qwen3-32b 88% 94% gpt-oss-120b 94% 91% The small model isn't

Apple primarily made the case for an improved experience with its longstanding Siri assistant, which like most other announcements had a hefty helping of AI.

Anthropic is releasing Claude Fable 5, its first Mythos-class model available to the public. The model comes with guardrails that block responses in high-risk areas like cybersecurity and biology.

Anthropic is releasing Claude Fable 5, its first Mythos-class model available to the public. The model comes with guardrails that block responses in high-risk areas like cybersecurity and biology.

Anthropic just announced Claude Fable 5, a new AI model it said is the most powerful model it has ever made widely available. According to the company, Fable 5 "shows exceptional performance in software engineering, knowledge work, and vision," with its lead over other models growing as tasks become longer and more complex. Fable 5 […]

Meta won't say why or whether it's coming back.

Apple used to question whether generative AI-powered editing features were worth the risk of distorting our perceptions of the world. Now it seems Apple no longer believes that photos should accurately capture reality. At WWDC 2026, the company announced a host of new AI-powered photo editing tools. They give users effortless powers of manipulating images […]

With SpaceX, Anthropic, and OpenAI all eyeing massive public debuts, the tech industry may soon have a new class of corporate overlords — and a new acronym to match. Say goodbye to FAANG and hello to MANGOS.

Because a reality wasn't given to us and is not there; but we have to make it ourselves, if we want to be; and it will never be one for ever, but constant and infinitely changeable. Luigi Pirandello, One, No One, and One Hundred Thousand Pirandello wrote this about the human condition. He didn't know he was describing the future of the internet. The web we know is about to disappear Not slowly. Not gradually. The web page, as the default unit of human navigation, is about to disappear: it will strip itself of everything we call "interface" and what remains will be only what it always was underneath — data, structure, instruction. The enticing homepages. The banners. The product carousels engineered by UX teams to capture attention in the first second and a half. The brand colors. The call-to-action buttons optimized for conversion rate. All of this is designed for a human eye that navigates alone. That eye is about to delegate. The agent that browses for you Imagine you want to buy a pair of shoes. Today you open a browser, search, filter, compare, go back, reopen the tab you closed, forget what you were looking for, start again. In a few years — maybe less — you will tell the agent what you want. The agent will already know that you have wide feet, that you prefer leather to synthetic, that you're looking for something for a wedding in June but deep down you want something that works afterward too. It will know that today you're in a practical mood, not an aspirational one. That you've spent a lot this month. The agent won't open a homepage. It will query a data structure. It will receive prices, availability, variants, return policies. It will build for you — and only for you, and only in that moment — a presentation tailored to measure. Colors that belong to you. Texts that speak your language. Images generated for your aesthetic sensibility of that day. The same store. Five billion different versions. One for each person, one for each moment. One, No One, and On

Custom agents let GitHub Copilot CLI understand your stack and team workflows, turning one-off terminal prompts into repeatable, reviewable processes. The post From one-off prompts to workflows: How to use custom agents in GitHub Copilot CLI appeared first on The GitHub Blog .

The Problem I was using Claude Code, Codex, and Cursor daily but had no idea how much I was spending on tokens. Bills kept surprising me. The Solution I built AIUsage — a local-first, open-source CLI that tracks everything. Key Features Token usage tracking with daily breakdowns Cost estimation with configurable pricing Model usage ranking Multi-device sync via GitHub or S3 Desktop widget How It Works bash npm install -g @juliantanx/aiusage aiusage parse aiusage serve Why Local-First? Your data never leaves your machine. No accounts, no API keys, no cloud servers. Try It [aiusage.jtanx.com](https://aiusage.jtanx.com)

A useful technical idea, repeated often enough, eventually generates an unuseful philosophical claim. The current example is grammar-constrained decoding. The technique is straightforward — at each generation step, the language model's next-token logits are masked so that only tokens whose continuation can satisfy a formal grammar remain selectable; the output is, by construction, structurally valid. JSON parses. SQL is well-formed. Function-call signatures match. There is a real engineering payoff and a healthy ecosystem of libraries that deliver it. The drift is not in the engineering. It is in the rhetorical move that follows the engineering. A growing corner of 2025-2026 AI writing argues, more or less explicitly, that constraining a model's output is making the model approach meaning — that filtering linear sequences is somehow building structure, and that structure is somehow building understanding. I want to take that drift seriously, because it is the same conflation Chomsky and collaborators flagged in their March 2023 essay in the New York Times , and the engineering literature on constrained decoding agrees with Chomsky on the substantive question, even when the marketing copy doesn't. What grammar-constrained decoding actually is A language model produces output one token at a time. At each step, the model emits a probability distribution over its vocabulary, and the decoding strategy (greedy, top-k, nucleus, etc.) picks one token. Without modification, the model is free to emit any continuation; the resulting text might happen to be valid JSON, or it might not. Grammar-constrained decoding intervenes in that step. A formal grammar — typically a context-free grammar, sometimes a regular expression, sometimes a JSON schema or Pydantic model — defines what counts as valid output. At each generation step, the constraint engine computes which next tokens could lead to a continuation that is still satisfiable under the grammar, masks the logits for all other