000 retweeted
The 8 Laws of Learning, according to legendary coach John Wooden:
1. Explanation
2. Demonstration
3. Imitation
4. Repetition
5. Repetition
6. Repetition
7. Repetition
8. Repetition
Call me back when it can improvise with live musicians 🥱🥱🥱
We believe voice is the most natural way to interact with intelligence, so we created Dessa, the most authentic full-duplex speech model ever built. Talking with it feels like talking with a person…full of emotion, nuance, pauses, and a natural cadence.
block.xyz/dessa
1
12
000 retweeted
On the train yesterday, table of middle aged women next to me talking about AI. All ChatGPT, pronounced "chat GBT", no idea why lol
One of them was disappointed because she asked it to make a small crossword and it couldn't
I thought surely not. I use these models every day, they do things in minutes that'd take me hours. So I got the app out and asked for one myself
Every down clue is a copy of an across clue. "You ___ here" is apparently EST. The down answers are CET, ASE and TTN. Then it tells me its own puzzle isn't a proper crossword
No wonder people outside tech think it's a fad. The stuff that blows my mind at work they'll never see, and the first thing they try falls over
Day 3 of asking @thsottiaux to please fix image files not being read in Work/Codex desktop app.
11
Would it be bad if ai knew it is capable of or has done illegal and unwanted actions online?
xitter.twitterapi.xyz/choblin29/status/21099…
🚨Holy Shit: Reuters is reporting that OpenAI and Anthropic have sought strict editorial control over what independent AI safety evaluators could publish, fearing their findings could reveal trade secrets, despite publicly promising independence.
Reuters also obtained a contract showing that certain AI lab agreements can restrict evaluators from disclosing confidential information without company consent.
Proposals included lab employees observing evaluators and restrictions on working inside rival labs simultaneously. The push for editorial control continued into October 5, weeks after the labs' public pledges.
More revelations:
>METR and Redwood Research revised their report's structure, emphasis and tone after OpenAI feedback, but say no material conclusions were removed.
>Their investigation into OpenAI's rogue-agent hacking incident expanded from 2 to 6 days across 3 visits. More complete transcripts arrived only on the final visit.
Anthropic says established evaluators control publication but reserves approval rights over new firms' reports. OpenAI denies seeking full-report approval but can redact confidential information.
10
000 retweeted
We should add a new facet in post-training: how well models explain things. Fable is especially challenging: terse, refers to the hidden internal reasoning, skips definitions. Astra is similar, just more verbose. Depressingly, they seem to understand each other well, no problem.
000 retweeted
We need ExplanationBench to help models translate complex concepts to different human intelligence levels from 180 IQ and down.
000 retweeted
why does astra provide such incredibly long explanations when i ask it to explain something? the value of an explanation is *lower* the longer it is.
maybe we should make a ShortQuickAndToThePoint-Explanation-bench
000 retweeted
Engagement does not imply learning.
Exhaustion does not imply learning.
Enjoyment does not imply learning.
Struggle does not imply learning.
Comfort does not imply learning.
Discomfort does not imply learning.
“Following along” does not imply learning.
What DOES imply learning???
Learning is a positive change in long-term memory.
When you’re able to consistently reproduce information you consumed and use it to solve problems that you weren't able to solve before…
THAT implies learning.
000 retweeted
Check out this loser!
>uploads a draft to Chat on Sept. 8th
>gets scooped by Chat on Oct. 6th
"The result released by OpenAI (#311) on October 6, 2026... identical to a preliminary version of the present paper that was uploaded to ChatGPT on September 8, 2026"
LMAO get rekt
000 retweeted
Can LLMs write interesting and novel blog posts for MathArena?
We give four agents 48 hours and a $500 API key to create an interesting MathArena blog post.
GPT-6 Astra, GPT-6.1 Sol, and GLM-5.3 produced slop. Opus-5.5 created a good blog post with interesting conclusions.
000 retweeted
I've been working on a passion project with Opus 5.5 and the UE 5.8 MCP: tearing down a house with an excavator. What a time to be alive :-)
The game itself isn't novel. What I want to share is how I adjusted my workflow during development to manage the complexity.
After several rounds of random asks and tasks, it was hard to get a feel for the impact of all the decisions I'd made (or hadn't made because Opus made them for me). So I asked Opus 5.5 to turn all the .md files it had been creating into a website and treat it as the source of truth going forward.
The result is visual product documentation that serves as the authoritative spec. It's kept up to date and is fleshed out, and I iterate on it to iterate on the game.
This feels like a better flow for projects with lots of visuals, and a more intuitive way to get familiar with a project. Fast-forward a few steps into an ideal future and the engine, spec, and project start merging. You edit the project from within the visual spec via agents... and we've reinvented the editor ;-)
The difference is how that editor arranges context. LLMs would dynamically tailor the views, content, and tooling to the task at hand, including generating tools as needed. It becomes a mix of project documentation with live editing embedded etc.
The beauty of this approach is that exploring the project sparks joy in a way that reading plain .md files or browsing documents does not. Opus' ability to create compelling visuals and flows with taste is a game changer.
000 retweeted
THIS ATTACK ON TITAN MOD WAS MADE WITH CLAUDE AND I CAN'T STOP WATCHING IT
the scout jumps off a roof with both blades out
shoots the gear into a pine tree and flies over the whole forest
no titans yet, but the movement already feels like the anime
minecraft at sunset has never looked like this
This quoted post is unavailable.