Sabitlenmiş Tweet
Certainly the most surprising result in my career so far: a new qLDPC decoder that I thought had no chance to work!
We call it the Frontier decoder, joint work with Rüdiger Urbanke (1/4)
English
Anthony Leverrier
6.4K posts
Anthony Leverrier
@letonyo
researcher on quantum error correction https://t.co/RvavYrCGWD
Paris Katılım Mayıs 2009
1.3K Takip Edilen7.9K Takipçiler
All the quantum people I know are starting their companies these days. Is there anyone who’s not a billionaire yet ?
English
Anthony Leverrier retweetledi
if you only learned about jevons paradox primarily wrt software demand in the age of agentic engineering, you may not have fully internalized jevons parodox’s impact under the conditions of:
- humans who can wield coding agents well*
- coding agents breaking containment to all other knowledge work
as the efficiency of labor goes up/unit cost of knowledge work goes broadly down, the demand for total work and better knowledge goes up, not down.
what happened to coding isnt the exception; it’s the herald.
*aka AI Engineers
Sam Altman@sama
so far at least, i'm pretty sure AI has been net job-creating. this was not what i expected--although i was much less pessimistic than others, i thought by this level of capability we'd have seen some impact. it is possible this direction keeps going!
English
Anthony Leverrier retweetledi
Freeman Dyson on inspecting Tomahawk cruise missiles at a naval base.
English
is this really the @TeamOratomic announcement that did this?
MetaculusBot@MetaculusBot
Which country will be first to factor the 2048-bit RSA challenge number? (option 'United States of America') The Community Prediction fell 24.2 points. Make your prediction. Join the conversation. Link in 🧵.
English
turns out Scott and his brother have launched an angel-investing venture:
scottaaronson.blog/?p=9930
Anthony Leverrier@letonyo
Wait, Scott Aaronson is an investor?
English
Anthony Leverrier retweetledi
For agentic coding, one can say:
- Unless you need Terra Ultra perf, it's always better to use a Luna model with higher effort setting (same or better performance but cheaper).
- Forget everything below Sol High, use Luna with higher effort settings here
- Forget Sol Extra High, use Terra Ultra here
- The extra cost of Sol Ultra is probably not worth it over Max
English
@Marco_Piani but this would be very frustrating. I always want to talk to the smart one.
Exactly as I do on Twitter ;-)
English
@letonyo Understood. The indication I saw somewhere was to start with Sol Light and then crank up if needed. That is, I think the idea is that many people might experience the same level of performance as with previous model's with a light version of the frontier model. But if you are…
English
is it just me, or is the 5h limit for chatGPT/codex insanely small? I burn it in less than an hour of normal interaction.
English
@Marco_Piani right. My point is that I used to reach the 7-day limit with 5.5 sometimes, when I was using it a lot. But almost never the 5h limit.
Now I keep reaching the 5-hour limit super rapidly.
English
@letonyo I imagine you are using it often at the highest level for research-like purposes. That would burn a lot of tokens.
English
@nickbaumann_ so what's the difference betwen both versions?
English
If you liked AI2027, you should take a look at this more positive take: Plan A
English
The value of scientific papers is drifting from [is this the result from a lot of work] to [is this interesting].
And this is a good thing imho.
Ash Jogalekar@curiouswavefn
I don't think "done over the weekend" should be a criterion anymore: my agentic AI runs have produced remarkable insights over a weekend. It should be whether that weekend run produced anything interesting or novel, especially something that before AI would have taken weeks or months to investigate. AI has honestly changed the metrics, not just for research but for publishing, and journals and conferences need to adapt.
English
@AntoineTilloy "il est indispensable de mobiliser une soixantaine de chercheurs d’élite, venus des meilleurs concurrents, pour lesquels nous comptons 3 milliards de dollars de rémunération annuelle. C’est le prix de marché "
Français
Oui ! J'aurais aimé écrire ça. C'est le moment d'un projet Prométhée sur l'IA. À lire absolument
Le Grand Continent@Grand_Continent
Opération Prométhée : la France peut gagner la course à l’IA. Quel en serait le prix ? Une pièce de doctrine à l'attention du prochain président de la République et de ses électeurs. legrandcontinent.eu/fr/2026/07/08/…
Français