Sabitlenmiş Tweet
Christian S. Perone
9.9K posts
Christian S. Perone
@tarantulae
Machine Learning, Computer Science, Math. Working with robots that drive London/UK.
London, United Kingdom Katılım Şubat 2009
1.5K Takip Edilen7.5K Takipçiler
Christian S. Perone retweetledi
New Datology Research: We expose "The Finetuner's Fallacy"
The standard approach to domain adaptation (pretrain on web data, finetune on your data) is leaving performance on the table.
Mixing just 1-5% domain data into pretraining, then finetuning, produces a strictly better model:
◾ 1.75x fewer tokens to reach the same domain loss
◾ 1B SPT model outperforms a 3B finetuned-only model
◾ +6pts MATH accuracy at 200B pretraining tokens
◾ Less forgetting of general knowledge
Tested across chemistry, symbolic music, and formal math proofs. SPT wins on every metric.
Led by @_christinabaek and @pratyushmaini, with the full Datology team.
English
@grok @arnau1700 @PolitlcsUK There you are, @PolitlcsUK spreading misinformation as usual. The case is NOT from London, it is from Folkestone as many other serious news reported.
English
The baby isn't from London - she's nine-month-old Nala-Rose Fletcher from Folkestone in Kent. She fell ill on 4 March (before the Canterbury cluster) and was transferred to Evelina Children's Hospital in Lambeth, central London, for intensive care. Tests confirm it's the same MenB strain as the Kent outbreak, though her family say she hadn't visited Canterbury.
English
🚨 UPDATE: A baby girl is the first London case of meningitis B and is currently being treated in intensive care at Evelina Children’s Hospital
She is from Kent and fell ill on March 4 - just days before a cluster of cases emerged in Canterbury
English
@michalwols Yes we both agree on that, it is just that there is a trend that seems to be very simplistic in how they are seeing data as if data alone is the solution to everything.
English
@tarantulae Yes for sure, just like full attention, both let us train larger higher capacity models
I'm just saying I favor using simple reliable models like resnets over things like inception as long as they keep improving with more data
English
Just a reminder that a lot of ppl used to say that data was the key for ML before we had transformers. It is easy to say that data is very important after you spent years figuring out architecture, training, massive infra issues, etc. Nothing says that this won't change again.
English
@michalwols I think that residual connections are a case that contradicts this for example.
English
@tarantulae so it only makes sense to add priors and tweak the algorithms when data is hard to get
English
@michalwols It is mainly to the trend of "data is the only thing that matters" that we are seeing lately. It is not. Note that *there was a time* where people tried to just add a lot of data, it just did not work without many architectural innovations.
English
@tarantulae what is this in response to?
there will never be a time where adding more (quality) data doesn't improve things
English
@PolitlcsUK Can you please review the news article from KentOnline, it is clearly saying she is from Folkestone and NOT from London. I think you should be careful with spreading misinformation during an outbreak, during COVID times we saw how damaging this can be.
English
Christian S. Perone retweetledi
Over 700,000 repos ship crypto libraries that default to a static IV, creating widespread key reuse. We also released mquire, a Linux memory forensics tool, and added 12 new open-source Claude Code skills for security engineering. March Tribune: mailchi.mp/trailofbits/ma…
English
90% of articles about AI now are either technophilia (uncritical love of AI and technology) or technophobia (fear or rejection of AI and technology). But this is a cultural problem much older than deep learning, the only solution is: education.
Noah Smith 🐇🇺🇸🇺🇦🇹🇼@Noahpinion
Two groups of people have been extraordinarily bad for the AI safety discourse: 1. Lefties like Emily Bender who dismiss AI as a useless toy 2. Doomers like Yudkowsky who force the discussion to only focus on superintelligence
English
Christian S. Perone retweetledi
London has incredible talent & entrepreneurial spirit. Thrilled to deepen @GoogleDeepMind’s roots here with our spectacular new building Platform 37 - a nod to AlphaGo’s legendary Move 37. It’s a tribute to Science & AI, and an inspirational space for our next big breakthroughs!
English
@HanchungLee I mean that FBI must have got information from somewhere (obviously, assuming that it is not an op). You can fit many Shaheds on a truck, enough to make much more damage than a truck with small drones.
English
@tarantulae neither fbi nor california police has jurisdiction in canada or mexico. border control is fed.
so yeah. also, iranian drones is huge vs those repurposed consumer drones.
English
this is so delusional that no llm can hallucinate this.
ABC News@ABC
BREAKING: The FBI has warned police departments in California that Iran wants to retaliate for American attacks by launching offensive drones against the West Coast, according to an alert reviewed by ABC News. abcnews.com/US/fbi-warns-i…
English
@HanchungLee I think that the concern is if they smuggled drones to Mexico a long time ago already. Take Ukraine's attack on Russia for ex., they smuggled a lot of drones in trucks, a domestic attack with smuggled material. Don't think it is probable as well, but no sure which info they got.
English
Christian S. Perone retweetledi
Happy to share new progress in AI for Maths @GoogleDeepMind .
In extremal combinatorics, AlphaEvolve has helped establish new lower bounds for FIVE classical Ramsey numbers - a problem so challenging that even Erdős commented on its difficulty.
Historically, computationally deriving these bounds required bespoke, human-designed search algorithms. For many of these bounds, the best previous results are at least a decade old. AlphaEvolve changes this by acting as a single meta-algorithm that automatically discovers the search procedures needed to find these new bounds. 📷
English
RT @adityakusupati: "Omni-modal" to help you find information even better! -- of course equipped with 🪆.
great work by the team in leverag…
English
infonce doing its magic
Google for Developers@googledevs
Announcing Gemini Embedding 2 ✨ the first fully multimodal embedding model built on the Gemini architecture. Now available in preview via the Gemini API and Vertex AI. The new model provides semantic understanding across 100+ languages — and support for modalities across text, images, video, audio and documents (PDFs) in a shared vector space. Start building today. Read the blog to learn more: goo.gle/4los3aI
English
Christian S. Perone retweetledi
Unveiling our new startup Advanced Machine Intelligence (AMI Labs).
We just completed our seed round: $1.03B / 890M€, one the largest seeds ever, probably the largest for a European company.
We're hiring!
[the background image is the Veil Nebula - a picture I took from my backyard, most appropriate for an unveiling]
More details here:
techcrunch.com/2026/03/09/yan…
@
Advanced Machine Intelligence (AMI) is building a new breed of AI systems that understand the world, have persistent memory, can reason and plan, and are controllable and safe. We’ve raised a $1.03B (~€890M) round from global investors who believe in our vision of universally
English