Dòng tin

RT by @jeremyphoward: Codex just found a “workaround” of not having sudo on my pc…

Has @AnthropicAI completely given up on making API usage reasonably-priced? Following the token-usage changes recently they announced various updates to *subscription* usage to make it more reasonable. But they've done NOTHING for API users. The cost is insane at this point.

R to @jeremyphoward: I've largely switched over to using GPT-5.5 in recent weeks, which I like nearly as much as Opus 4.6 and 4.7, and is *very* reasonably priced (since I can use my subscription with the @OpenAI API.)

R to @jeremyphoward: (This is using our in-house harness, Solveit.)

Worked on some code this morning using Opus 4.8 and so far I'm really liking it. Much more cooperative than 4.7 and less "over agentic". Stops and asks for my input when needed in places 4.7 (and GPT 5.5) would just foolishly blast ahead. (Still WAY too expensive.)

AnswerDotAI released 0.14.2 at AnswerDotAI/fasthtml

AnswerDotAI released 0.1.3 at AnswerDotAI/fastaudit

RT by @jeremyphoward: My MLSys keynote on AI writing systems code got more interest than I expected. The recording will take a while, so in the finest tradition of AI labs sharing blog posts, we’re starting the Core Automation Blog with this one https://www.coreauto.com/blog/when-ai-starts-writing-systems-code

RT by @jeremyphoward: wait… if most people think 5.5 is better than 4.7, i assume that’s due to terminal coding benchmark… 4.8 is still outperformed by 5.5

RT by @jeremyphoward: Fascinating results + Anthropic running away with it right now + So many people want to start their own company + Google over OpenAI + Vercel, Linear, Every, PostHog overperforming A great list if you're trying to figure out where to go work 👇

RT by @jeremyphoward: $65B private round More than double the size of the largest IPO ever

RT by @jeremyphoward: glad to know Mythos' safety concerns have been addressed right as Anthropic also secured tens of billions in inference compute 👍

RT by @jeremyphoward: Fun fact: Australia is basically Scandinavia with the Sahara Desert bolted on.

RT by @jeremyphoward: There is a lot being written about the stylistic tells of AI writing (em-dashes, etc.) but this paper looks at AI narrative tells Fascinating differences between AI & human narrative, and asking AI to write in different styles doesn't do much to change it https://arxiv.org/abs/2604.03136

RT by @jeremyphoward: Anthropic is too expensive and will either lose customers or cut prices

RT by @jeremyphoward: For over a decade, we’ve accepted that end-to-end backprop is the only way to train deep networks. But holding the entire network in memory all at once is why AI training is hitting a resource wall. We found a new way to break the network into blocks and train them independently. The trick? Treating the network’s forward pass like a diffusion model denoising a signal. This reinterpretation slashes the memory needed to train deep models. In our #ICLR2026 paper (https://arxiv.org/abs/2506.14202), we matched end-to-end performance across ViTs, DiTs, and LLMs. We did this while training just one isolated block at a time.

RT by @jeremyphoward: Behind the MiMo API Price Reduction: The deepest price cut, up to 99%, is for Input (Cache Hit). The core reason is our inference framework now supports hierarchical KV cache optimization for SWA. Production inference engine tests show this optimization increases cached token capacity by 5x, equivalent to an 80% reduction in caching costs. Combined with Cache Read Overlap among multiple Full Attention modules in the Hybrid model, actual costs are further reduced. Prices for Input (Cache Miss) and Output are also reduced by 60%-80%. This mainly benefits from the extreme 1:7 Full:SWA sparsity ratio brought by the model architecture (the prefill compute of the 70-layer MiMo-V2.5-Pro roughly equals a 10-layer GQA model). This kept our original inference costs well below the industry average, naturally leaving a 2x-3x profit margin in pricing. This price adjustment simply reflects our decision to pass these structural cost efficiencies directly to developers. Operating at these newly reduced API prices, our production inference engine is running at near full capacity, and we can still essentially break even. We previously advised LLM companies not to "blindly cut prices" precisely because very few model architectures and inference optimizations can keep API costs from running at a loss. If more architectures that save compute and KV cache emerge, along with better inference Infra to drive down API costs, this will form an excellent virtuous cycle in the industry. More crucially, affordable, high-performance model APIs will drive real, sustained, and at-scale inference demand. This upstream demand pulls forward the development of the entire AI infrastructure chain—including chips, servers, optical transceivers, PCBs, liquid cooling, power, energy storage, and data centers—serving as a strategic fulcrum for a systemic revaluation of AI hardware. In the long run, this injects more affordable and accessible compute into both training and inference pipelines, accelerating the parallel evolution of global AGI across multiple regions and technical routes. For more technical details, we will release a detailed Blog post later.

R to @jeremyphoward: Impressive caching too...

Wow. It looks like the @XiaomiMiMo v2.5 model is insanely good value :O (Price for each prompt shown after each answer. Context includes >40k tool descriptions, system prompt, skills, etc.)

RT by @jeremyphoward: recommended reading. i too am very done with people anthropomorphizing a bunch of matrices on a GPU cluster, especially if the same people do not give two fucks about actual human beings.

RT by @jeremyphoward: the models from 6 months ago kinda feel the same like the recently released models. currently not holding my breath for more step changes, but will be happy if they happen.

RT by @jeremyphoward: Nothing disappoints me more than people saying we should stop progress because peoples meaning depends on that monotonous labor, as if humanitys highest purpose is filling Excel sheets or stocking shelves. This is the worst take. Worse than keep4o, anti ai art, doomerism,...

RT by @jeremyphoward: This from @TuhinChakr is brilliant. That prize winning story from Granta? Turns out it's just a bunch of random whole phrases taken directly from existing text on the internet. Tool allows you to trace those n-grams directly to their source, which is mostly random fanfiction. https://tuhinchakrabarty.substack.com/p/ai-slop-grantagate-and-bad-writing

I Don’t Want a Learning Dashboard for My Child

Breaking the Spell of Vibe Coding

How To Use AI for the Ancient Art of Close Reading

Stop Saying Boredom is Good for Kids

A Guide to Solveit Features