AI enthusiast passionate about research and bridging the Arabic gap in artificial intelligence.
و إن شاء الله هعملهاlinkedin.com/in/dsomarkhale… The Universe Joined September 2023
This one paper might kill the AI scaling hype.
While Big Tech burns billions on massive datasets, researchers just achieved state-of-the-art agent performance using 78 samples.
And it makes a scary amount of sense.
Here's the full breakdown:
بفتكر ايام ثانوية عامة ( و اول سنتين فالكلية) لما كان جزء من المدرسين و معيدين الكلية يعاملوني معاملة غريبة فشخ و لما اسألهم يقولولي "هو كدا شكلك مش عاجبني"
لحد ما قررت استلسم و احلق
بفتكر ايام ثانوية عامة ( و اول سنتين فالكلية) لما كان جزء من المدرسين و معيدين الكلية يعاملوني معاملة غريبة فشخ و لما اسألهم يقولولي "هو كدا شكلك مش عاجبني"
لحد ما قررت استلسم و احلق
This Tencent paper shows a way to improve reasoning by training only on raw text using reinforcement learning.
It is called Reinforcement Learning on Pre-Training data (RLPT) and it removes the need for human labels.
Simple “predict the next segment” rewards are enough to…
DSPy and ColBERT are interesting academic experiments imo.
Each is a multi-paper repo that has one coherent artifact, combining our latest research together.
We typically release the features as open source—hence get users/feedback—well before writing a paper on the new ideas.
DSPy and ColBERT are interesting academic experiments imo.
Each is a multi-paper repo that has one coherent artifact, combining our latest research together.
We typically release the features as open source—hence get users/feedback—well before writing a paper on the new ideas.
Talking to grad students, too many think that long-term projects (not scattered papers), proper code releases, thoughtful benchmarks are "not incentivized".
Most often they're mistaken. If we're talking incentives, *nothing* matches demonstrating impact! Will blog on this soon.
Prediction: In ~3 years academia will be the most desirable place to do fundamental AI research
Contributing factors:
- small models improve/become significantly more impactful
- open weights community broadens its reach
- gpus continue to get faster & cheaper
- meaningful…
You're in a ML Engineer interview at Meta, and the interviewer asks:
"Why does RL work better than supervised learning for LLMs?"
Here's how you answer:
كنت بدوّر على كورسات عن إزاي تبني Start-up،
بحيث تكون فاهم إيه اللي هيتم وتوسع وجهة نظرك بشكل أكبر. وصلت للكورس ده فحبيت أشاركه معاكم، يمكن يساعد شخص يبدأ مسيرة مهنية جديدة بإذن الله ❤️
"It's autocomplete" is not a helpful analogy to understand LLMs. A LLM is more like a database that lets query information in natural language. You can query both knowledge, and "patterns" (associative programs seen in the training data, that can be applied to new inputs).
“The biggest lesson that can be read from 70 years of AI research is that general methods that leverage computation are ultimately the most effective, and by a large margin. The bitter lesson is that building in human knowledge is a losing game in the long run.” – Sutton
“The biggest lesson that can be read from 70 years of AI research is that general methods that leverage computation are ultimately the most effective, and by a large margin. The bitter lesson is that building in human knowledge is a losing game in the long run.” – Sutton
"We should stop trying to find simple ways to think about the contents of minds, such as simple ways to think about space, objects, multiple agents, or symmetries."
Richard Sutton
Most RL for LLMs involves only 1 step of RL. It’s a contextual bandit problem and there’s no covariate shift because the state (question, instruction) is given. This has many implications, eg DAgger becomes SFT, and it is trivial to design Expectation Maximisation (EM) maximum…
How does backprop work with RL?
The virtue of backprop is that it updates EACH individual parameter in proportion to how much wiggling it affects the loss. This is only possible if you know how changing each parameter affects the loss function.
But of course with RL this is…
92 Followers 900 FollowingExploration over Exploitation.
RA @Mila_Quebec, Research Fellow @UniofOxford. MSc @UWindsor. Interested in Adversarial attacks, security & reliability of LLMs
175 Followers 763 FollowingSr. WordPress & Shopify Developer | Build high-performing modern websites and E-commerce solutions—a big fan of Headless CMS, Gutenberg Block Development.
2K Followers 4K Following#Scientist #Researcher #Author of 20 research papers #Catalysis #Water splitting ; Content Creator; Professional adviser in paper writing ❤️ "single"
4K Followers 7K FollowingBest Online English Teacher that enjoys helping students pass the IELTS, OET, General English tests #ielts #esl #english BSc MSc CertTESOL
298 Followers 8K FollowingYes, I can see some risk that your threat to jail internet company executives for not censoring aggressively enough to backfire.
2K Followers 1K FollowingHi, I am Srishti.
Aspiring Data Scientist l soft tech soul.
I post about:
-data analytics and ML
-skincare + soft glow
- life while learning gently follow along
14K Followers 12K FollowingLoving Father, LUNC Lover, Pinball Collector and Enthusiast, DeFi, Crypto, Classic Rock, Movies, and Candy Cigarettes 🚬 . That about sums it up for me.
3K Followers 51 FollowingLearn how to build AI Agents & sell them to local businesses 💸 Founder of @getoutbox_ai Learn how to build AI Agents for FREE 👉 https://t.co/q9zPwllLOC
36K Followers 5K FollowingExperienced Data Science Leader | PhD in Machine Learning | 4x Author | Black Belt 🥋 in Time Series | Chief Conformal Prediction Promoter| Mathematician |
26K Followers 1K FollowingGenAI @Youtube | Building AI powered video editing | ex : @Google Search & @Microsoft Azure | 3x hackathon winner | Views my own
13K Followers 2K FollowingAssociate Professor at Harvard & Kempner Institute. Applying computational frameworks & ML to decode multi-scale neural processes. Marathoner. Rescue dog mom.
45K Followers 1K FollowingNeuroscientist interested in cognitive-emotional brain
Author of The Entangled Brain (MIT Press); The Cogitive-Emotional Brain
Neuroscience & Philosophy Salon
28K Followers 10K FollowingNeuroscience,Insular cortex, Neurobiology RT≠endors I do not reply to direct messages
mstdn: @[email protected]
Bluesky: @claeneuro.bsky.social
27K Followers 296 FollowingProfessor of linguistics and professor of computer science at Stanford and author of the James Beard award finalist "The Language of Food"
10K Followers 598 FollowingJanice M. Jenkins Collegiate Professor of Computer Science at U. Michigan, Director @Michigan_AI Lab, Former @ACLmeeting President, Researcher #NLProc #AI.
15K Followers 7K FollowingI build tough benchmarks for LMs and then I get the LMs to solve them. SWE-bench & SWE-agent. Postdoc @Princeton. PhD @nlpnoah @UW.
54K Followers 2K FollowingPioneers in generative AI. chat/image/video/music. you own outputs. $4.99/mo for DeepAI Pro
cool research at @arxiv_daily
For support email [email protected]
69K Followers 2K Following✨ AI should be about empowering humans, building understanding, and making dreams realities. 👩💻 DevX Eng. Lead @GoogleDeepMind ex-@GitHub || views = my own!
218K Followers 425 FollowingThe latest rumors and developments in the world of artificial intelligence. DM to include your AI project in the newsletter.
102K Followers 174 FollowingProfessor of computer science at UW and author of '2040' and 'The Master Algorithm'. Into machine learning, AI, and anything that makes me curious.