antonbuild retweeted
Pensée émue. Je souhaite un prompt rétablissement aux journalistes de France Info.
Le directeur de la rédaction de France Info informe avoir mis en place un accompagnement… médical pour les 2 journalistes qui ont interviewé Sarah Knafo :
"je ne veux pas surjouer la chose mais ce sont des moments compliqués à vivre…" franceinfo.fr/replay-radio/l…
antonbuild retweeted
1. Benchmarks provide one piece of evidence for acceleration. Since Oct 2025, AECI scores for Anthropic’s models have grown at 1.9x [95%: 1.5–2.8] the earlier rate. Analysis choices affect this estimate, and it remains unclear how benchmark scores map to real-world capabilities.
antonbuild retweeted
stealing content is the fastest way to get views for your product
I created an AI CMO which constantly scrolls TikTok to find the best slideshows to replicate to promote your startup
Creates the content
And publish it automatically on your accounts
It’s the easiest way to scale your product to millions of views
Building a company has never been easier
antonbuild retweeted
Introducing the Fusion API, the smartest compound model in the market.
Fusion achieves Fable-level intelligence at half the price.
How it works 👇
antonbuild retweeted
Opus 4.7 - 400k vs 1m context - is there a difference?
I've heard @theo talk about the fact that it is unlikely that Anthropic would have offered up a model with 1m context at the same cost, if it wasn't a different (i.e. cheaper to serve) model.
I did a test where I toggled the 1m default model on & off in Claude Code (otherwise default settings, xHigh reasoning) and compared the outputs with 3x generations - same prompts etc.
My observations:
- Models feel DIFFERENT - often when you ask a model for the same generation, you get a somewhat different answer, but it feels & smells the same. Here 400k and 1m are very different every time
- 400k model seems better - not that 1m is trash and 400k is amazing, but there are definitely issues with the level of ambition and accuracy that 1m model seems to have
Examples of 1m failing:
- Voxel Rome: the colosseum is nowhere near as impressive
- Golden Gate: cars go sideways, waves not very high, bridge goes into land; though the structure of the bridge is a bit better
- Stonehenge: structure is more 'wrong', lighting, shadows & textures are more flat and not as rich
This isn't a conclusive evidence of course, but at least to me the two models do not behave the same way. Anecdotally as well when building 1m felt like it was doing more weird validation (e.g. going around in circles) and 400k was more straightforward. These sorts of things are harder to capture in tests, but you'd notice in Claude Code.
You can review the hosted generations, see the code & prompts in the links below
Closed labs hide model sizes. They can't hide what their models know, and what a model knows is an indicator on how big it is.
Reasoning compresses. Factual knowledge doesn't. So you can size a frontier model from black-box API calls alone, and across releases you can literally watch a single fact arrive in the parameters over time.
For three years, my friends Jiyan He and Zihan Zheng have been asking frontier LLMs the same question: "what do you know about USTC Hackergame?", a CTF contest. May 2024: GPT-4o invented fake titles. Feb 2025: Claude 3.7 Sonnet listed 19 verified 2023 challenges. By April 2026, frontier models recall specific challenges across consecutive years.
After DeepSeek-V4 dropped, I instructed my agent to spend four days autonomously turning that habit into Incompressible Knowledge Probes (IKP) — 1,400 questions, 7 tiers of obscurity, 188 models, 27 vendors. Three findings:
1/ You can approximately size any black-box LLM from factual accuracy alone. Penalized accuracy is log-linear in log(params), R² = 0.917 on 89 open-weight models from 135M to 1.6T params. Project closed APIs onto the curve → GPT-5.5 ~9T, Claude Opus 4.7 ~4T, GPT-5.4 ~2.2T, Claude Sonnet 4.6 ~1.7T, Gemini 2.5 Pro ~1.2T (90% CI: 0.3-3x size).
2/ Citation count and h-index don't predict whether a frontier model recognizes a researcher. Two researchers with similar citation profiles get very different responses. Models memorize impact — work that shaped a field, not many incremental papers.
3/ Factual capacity doesn't compress over time. Across 96 open-weight models across 3 years, the IKP time coefficient is statistically zero, rejecting the Densing-Law prediction of +0.0117/month at p<10⁻¹⁵. Reasoning benchmarks saturate; factual capacity keeps scaling with parameters.
Website: 01.me/research/ikp/
Paper: arxiv.org/pdf/2604.24827
Anthropic said Mythos was too dangerous to release. Then four random guys in a Discord gained access on day one by guessing the URL...
This is pretty insane:
→ Group in a private Discord guessed the endpoint from Anthropic's naming conventions
→ They figured out the conventions from the leak in the Mercor breach three weeks ago
→ Used a contractor's legit eval credentials to walk in
→ Have been using it ever since to build simple websites
The AI that finds zero-days in every operating system on earth was defeated by address bar autocomplete... big yikes
Anthropic's Mythos has been accessed by a small group of unauthorized users, raising questions about control of the AI model bloomberg.com/news/articles/…
antonbuild retweeted
Replying to @JorgeA77832
I agree the spending is more modest as a share of US GDP at the time — here's the graph
antonbuild retweeted
My time tracking sheet where you can see what I am working on, in 15 min intervals
docs.google.com/spreadsheets…
antonbuild retweeted
Replying to @_mattwelter
not quite there though (text on first image + skin is way too perfect)
I guess you used gpt image gen 1 ? It has this same uncanny vibe
Try the same prompt with imagen 4, stuff like (boring reality pic, imperfect pic, bad shot, no composition, sharp background, … )
antonbuild retweeted
Replying to @ns123abc
honestly makes a lot of sense
OS allow user to interact with favorite apps
The clear OS winner is Apple because they built an ecosystem of hardware and software not just pure software
And Sam knows it
hence why Johny Ive is working with OpenAI to create a closed ecosystem with hardware… just like what he did at Apple
antonbuild retweeted
Replying to @pitdesi
history often rhymes
kodak invented digital photography then locked it in a drawer because selling it meant killing its own film biz
someone else shipped that tech and nuked kodak
google basically wrote the transformer playbook yet might be too scared to go all in
if they don’t fully lean into ai someone else will scoop them and steal their golden goose
antonbuild retweeted
Introducing The Garage
An industrial design space for tinkerers to wander, be creative and turn ideas to reality.
We believe the next set of physical innovations are going to be created here.
If you're building in hardware, reply with what you're making
antonbuild retweeted
I wrote an essay about Gmail’s useless email-writing AI assistant: “AI Horseless Carriages”.
link and TLDR in thread
antonbuild retweeted
before moving to SF
i assumed the people here knew what they were doing.
turns out no one knows anything.
no clue. no reason. no masterplan.
just the audacity to begin.
the future is rarely built by ppl with a plan.
antonbuild retweeted
NEW: Replit founding data scientist and engineer @giansegato: “I’ve begun to see product managers developing business financial models; designers writing commercial ads; barbershops building custom booking systems; and restaurant owners creating advanced pricing tools... These people always had it in them to do these things. The key difference is that it no longer takes years to learn how.”
Gian’s in Pirate Wires today with a piece on the "unraveling of credentialism" — and how having an edge in the market is no longer about knowing how to do one specific thing very well, but about being biased toward making it happen. Link threaded 👇
antonbuild retweeted
Today, we're officially launching @OriginsInc.
The podcast discovering the early, unheard stories of icons from all industries.


































