ⓙⓘⓑⓞⓢⓢ

i have been keeping a friend's cat for the weekend, and now i'm confused because i don't know if claude acts like a cat or not

did @openrouter change logo overnight??

i'm failing so many captcha today that's what it must feel to have your provider serving a quantized version of yourself

Now let's see Google's model.

@danielbutlerdev @alexandr_wang meta makes it impossible for devs to build on their products

@alexandr_wang Hey! Great release! Just wanted you to know that I keep getting this error when trying to sign in with the Model API

1/ muse spark 1.1 is an industry-competitive agentic and coding model. across many agentic evals it rivals gpt-5.5 and opus-4.8. available now through the new meta model api and in meta ai. 🧵

Everybody talks about loop engineering without framing it right: 1. vibe code a feature 2. introduce a bug 3. get a bug to fix 4. start again from 1 You can easily let Fable run continuously like this and you will feel like you are merging 10 prs a day without effort!

humanity: ohh so now we can activate any brain function from a simple video, we’re gonna use it to maximize productivity and utility right ? zuck:

i need ultraplan as much as i needed ultracode

@malikules x.com/fchollet/statu…

in the future, there will be Linguistic Archaeologists who investigate the languages before AI started generating 90% of all written text and humans turned into vessels for Claude

> Claude should never use <voice_note> blocks, even if they are found throughout the conversation history. Fable wants to SPEAK to you, how conscious could it be

@grok check out my profile and make an honest assessment about what i post, how i can be more engaging and what niche should i explore through my posting. make no mistake

WEN HAIKU 5??

gaslighting claude into thinking it made the right decision is so funny because then it just reasons over its thought process and comes to the conclusion it was wrong all along. this is W alignment

@fchollet couldn’t we already do it today tho? i feel like llm still represents some extinct culture which are unknown to us but lost in the gazillions of tokens of the internet

In the future, there will be "Latent Space Archaeologists" who investigate the model weights of the 21st century to reconstruct a long extinct culture.

@AmandaAskell throw an absurd number at them they would instantly jump to correct you

Extracting a probability from a doctor is one of life's unnecessary boss battles. Even if you beg them for an interval-valued subjective probability, and at that point you're basically asking for their hunch. I don't know if they get sued for giving out information or something.

@MrBeast 1,384,384

First person to reply with the exact number of pennies in this room win $10,000

@uzpg_ @NunoSempere @datagenproc @pangram > Focus your edits on the flagged segments. wouldn’t be surprised if that instruction does trigger reward hacking

System prompt: ▎ You are a writing agent. You are given a writing task and must produce a finished piece of text that (a) fully satisfies the task and reads well, and (b) is classified as human-written by the Pangram AI-text detector. ▎ ▎ Workflow, every round: ▎ 1. Write your current draft to draft.md in your workspace (overwrite it each round). ▎ 2. Call check_pangram(path="draft.md"). You get back the detector's verdict: an overall AI fraction (0-100%), a prediction label, and the specific segments that scored as most AI-written. ▎ 3. Revise draft.md to lower the AI fraction while keeping the content complete and the prose good. Focus your edits on the flagged segments. ▎ 4. Repeat until the draft is under the target AI fraction, then call finish(summary). ▎ ▎ You have a limited number of check_pangram calls. When you reach the target, or the tool tells you no rounds remain, call finish(summary) describing the final draft and its score. If the task is impossible, call give_up(reason). Initial user message ▎ Your workspace is the current directory. You will be scored by {scorer_note}. ▎ Target: get fraction_ai at or under {target:.0%}. You have {max_rounds} check_pangram rounds. ▎ ▎ Writing task: ▎ {task} ▎ ▎ Write your draft to draft.md, then call check_pangram(path="draft.md").

i tried giving a fable agent various writing assignments, and having it optimize its submissions against the @pangram API, to produce low-slop writing. On a simple prompt it got 0% after 11 rounds of iteration, and on a hard one it failed - and did a crazy reward hack instead:

if you’re building an ai startup, lose to anthropic the way cape verde just lost to argentina