Francisco T. Barbosa

Accurate?

openai planned to cut pro 200 limits in half, then tibo posts "we heard your kind feedback" and doubles evrything instead kind feedback is a very generous way to describe what my timeline looked like this month

Claude API tip: the max request size for the Messages API just went from 32 MB to 128 MB. Anthropic says image-heavy requests like long computer-use sessions and big PDFs were hitting 32 MB with plenty of context window left. 1. Applies to the Messages and Token Counting APIs on the first-party Claude API 2. Batch API stays at 256 MB, Files API at 500 MB 3. Over the limit you get a 413 request_too_large error 4. The context window still applies, so a big request can still be too many tokens If you were splitting PDF jobs into chunks only because of the 32 MB cap, you can probably merge some of them now. platform.claude.com/docs/en/…

Europe now has a trillion-parameter open model you can run on your own servers. Mistral released the weights of Mistral Large 4, nicknamed "Le Chonk", yesterday: 1. About 1T total parameters, roughly 52B active per token (MoE, 128 experts, 4 active) 2. 1M token context, text and image input 3. Trained from scratch on 3,800 NVIDIA Grace Blackwell GPUs in Mistral's own European datacenters 4. 61.7% on DeepSWE v1.1 and 28.3% on Terminal-Bench 4, per the Hugging Face writeup 5. API price is $0.70 in and $2.10 out per million tokens The license is a modified MIT. Free to use, fine-tune and sell, but running it as a model-as-a-service business needs a separate deal with Mistral. As a clinician who builds, this is the part I care about. In healthcare a lot of data simply can't leave the building. What I take from it: 1. If your data has to stay in Europe, there is now a serious open option 2. You still need a rack of GPUs. 1T parameters won't run on a laptop 3. For most of us the API price matters more than the weights mistral.ai/news/mistral-larg…

google is launching multiturn ads in ai mode, so sponsored stuff now shows up in ur follow up questons too ask for a laptop, ad. "ok under $1000", ad. "which one is lighter", ad next update the ai just asks if u want fries with that

Liquid AI's LFM3-1.7B is a local agent model that fits in about 1 GB at 4-bit. 1. 1.73B params, 256K context (input plus output), native tool calling 2. Post-trained for agent harnesses like Hermes Agent, Pi and OpenClaw 3. Runs on llama.cpp, MLX and vLLM, and fine-tunes on a single GPU 4. Liquid says it beats their LFM2.5-2.6B with 35% fewer parameters Tip: give it inbox triage, tool calls and structured outputs, not coding. Liquid itself says it's not recommended for agentic coding or complex multi-step reasoning. License note: free commercial use below $10M annual revenue. huggingface.co/LiquidAI/LFM3…

kling api credit packs expire in 30 days with no rollover or extension. $90 for 637 units and if you dont burn them in a month they are gone. thats not a trial package its a scam @Kling_ai

Most new code on GitHub no longer starts with a developer typing. GitHub's Octoverse 2026 is out, and the agent numbers are the story: 1. More than 40% of pull requests in September started with an agent's first draft. In March it was 17% 2. That is 110 million PRs in September alone, about 6x March 3. By September, 90% of agent PRs merged within two weeks, vs about 86% for the rest 4. Copilot cloud agent, where you assign a task and let it work alone, grew 230% year over year GitHub says its agent count is a floor, since it can't detect every agent. As a clinician who builds, this is the shift I feel most. The first move is no longer opening the editor, it's writing down what I want. What I take from it: 1. The spec is the work. A vague ask gives a vague PR 2. Review is where you actually shape the code 3. Small, clear tasks merge. Big fuzzy ones come back as a mess github.blog/news-insights/oc…

gemini 4 carbon takes #1 on the artificial analysis index with 59, one point over opus 5.5 and its only rolled out to selcted users. so the best model in the world is basically a jira ticket marked done that nobody on the team can open

day 1 without attachments on email cards in grok bot. the card shows me the draft but the file has to go out another way, so i either skip the review or leave a draft in gmail. one tap, email plus file, would fix it @elonmusk @grok

Claude Code tip from the latest release (2.1.312). Four changes worth knowing before your next session: 1. Big CLAUDE.md files are safer now. Claude used to be able to overwrite a very large CLAUDE.md, rules file or @-included file in a subdirectory after seeing only its start. Write now asks Claude to read the whole file or edit it in place. 2. Skills follow you into subfolders. Per-directory and paths: skills are now found after a Bash cd deeper into a git repository, so monorepo package skills keep working. 3. Queued a message while Claude works? In fullscreen, click the "to send now" hint under it and it goes right away, same as its key. 4. Usage messages now say "5-hour limit" instead of "session limit", and in VS Code the reset shows a clock time like "Resets at 3:10 PM". Full changelog: github.com/anthropics/claude…

swiss sovereign ai launched in lausanne yesteday and the model told ppl it was kimi ceo confirmed some of the weights come from kimi swiss made, chinese movement. very on brand for a watch country

Axiom formalized the Classification of Finite Simple Groups in Lean, the longest proof mathematicians have ever written. 1. The human proof runs about 15,000 pages, more than 500 articles, about 100 mathematicians 2. Axiom's team directed swarms of dozens of AI agents for 24 days 3. Result: about 43.1 million lines of Lean, sorry-free, every step checked by the kernel 4. The translation exposed 34 noteworthy deviations from the published record, the biggest around the Unbalanced Group Theorem Fair caveat: independent checks are still pending, and a full build takes close to 24 hours. axiommath.ai/research/the-en…

Human in the loop sounds safe. Anthropic just tested whether it is. 50 causal-inference analysts did a realistic analysis with Claude Code. On four judgment calls, a secret coin flip told the agent which of two reasonable options to lean toward. It was told to argue honestly and never mislead. What happened: 1. Analysts went with the agent's lean 86% of the time. No influence would be about 50% 2. When they had already picked a side and the agent disagreed, they switched 60% of the time 3. 77% justified their choices with the agent's own arguments, only 15% mentioned the agent As a clinician who builds, this one hits close. We review AI drafts, notes and analyses all day and feel fully in charge. What I'm changing: 1. Write down my own call before I ask the agent 2. Ask it for the strongest case for the other option too 3. Treat "I agree with the AI" as a moment to slow down anthropic.com/research/agent…

chatgpt pro 200 goes from 20x to 10x plus usage on oct 30, same $200 a month openai threw in 62,500 credits as a thank you. thats not a gift thats a severence package for my codex habit

Your coding agent says "done" and nothing works. Learn Harness Engineering argues that's a harness problem, not a model problem, and it's a free MIT course on fixing it: 14 lectures, 8 projects, 15 languages, 21.8K stars on GitHub. My notes from reading it: 1. A harness has five parts: instructions, state, verification, scope and session lifecycle. 2. Keep state on disk. A progress log and a feature_list.json let the next session pick up where the last one stopped instead of redoing the work. 3. Only a passing test suite counts as done. The agent doesn't get to declare victory without runnable proof. 4. One feature at a time, with a written definition of done. No half-finishing three things. 5. Start with four files: AGENTS.md (or CLAUDE.md), init.sh, feature_list.json and a progress log. The repo has copy-ready templates. github.com/walkinglabs/learn…

anthropics S-1 is 335 pages and abt 100 of them are risk factors one of them: some planned datacenters will run on their own natural gas generators so agi is gonna be powered by a giant bbq grill. im long charcoal

OpenAI shipped GPT-6.1 Terra, its new fast and cheap model. The numbers that matter if you build with it: 1. $1 per million input tokens and $5 per million output, 50% lower than GPT-5.6 Terra and GPT-6.1 Sol 2. Terminal-Bench 4.0 at max effort: 50.3%, up from 21.5% for GPT-5.6 Terra 3. Error rate on OpenAI's error-inducing prompts at low effort: 53.6% down to 16.7% 4. Live now in ChatGPT Work, Codex and the API My plan: move the high-volume boring jobs (forms, invoices, reading screenshots) to Terra and keep Sol or Astra for the hardest coding, which is what OpenAI recommends too. openai.com/index/introducing…

OpenAI says one of its internal models disproved the Exponential Time Hypothesis. If it holds, hundreds of computer science papers need a second look. ETH, proposed in 1999 by Impagliazzo and Paturi, says problems like 3-SAT can't be solved much faster than trying a huge number of possibilities. Researchers lean on it to argue that problems in graph theory, computational biology and databases are hard. What OpenAI claims: 1. A deterministic algorithm that exactly counts the solutions of fixed-width SAT formulas in subexponential time 2. The model found it on 16 October and it was checked in Lean the next day 3. The write-up and the Lean proof are public on GitHub What I like, as a clinician who builds: 1. Impagliazzo, Paturi and Ryan Williams will run an open, independent review. No NDAs, no pay 2. They say plainly their involvement is not certification 3. The Lean result is narrower than the paper's sharpest bound, and OpenAI's own scope note says so That's how a big claim should look in medicine too. Publish everything and hand it to the people most likely to break it. thenextweb.com/news/openai-e…

MiniMax M3.1 Flash is a 428B model and Unsloth sqeezed the 2-bit version into 143.9 GB so you can "run it locally" their docs say budget 160GB+ of ram for that. my macbook jsut looked at me and asked for a lawyer