Sabitlenmiş Tweet
Interested in language, knowledge and reasoning? Come to work with me and others at @DeepMind and its language group as a research scientist! Apply here: boards.greenhouse.io/deepmind/jobs/…
English
Sebastian Riedel (@[email protected])
1.9K posts
Sebastian Riedel (@[email protected])
@riedelcastro
Researcher in NLP/ML @deepmind, @ucl_nlp, @[email protected] on Mastodon
London, England Katılım Eylül 2009
456 Takip Edilen16.4K Takipçiler
Sebastian Riedel (@[email protected]) retweetledi
🚨 New Paper 🧵
How effectively do reasoning models reevaluate their thought? We find that:
- Models excel at identifying unhelpful thoughts but struggle to recover from them
- Smaller models can be more robust
- Self-reevaluation ability is far from true meta-cognitive awareness
English
Sebastian Riedel (@[email protected]) retweetledi
Want to build on the new @Google Multimodal Live API with Gemini 2.0? My teammates @hapticdata + @tinaz0ne + @trudypainter made open-source starter demos! 🧵
Here's a React.js boilerplate console you can start with. Code here: github.com/google-gemini/…
English
Sebastian Riedel (@[email protected]) retweetledi
The Gemini 2.0 era begins with 2.0 Flash Experimental release ⚡️
📈2.0 Flash beats 1.5 Pro across factuality, reasoning, coding, math.
📳 More modalities - image and audio out (in EAP)
🔧 Native tool use for Google Search, code execution and 3P functions
🆕 a new multimodal, realtime API experience
🎬 3 new cool starter apps - spatial understanding, video analyzer and map explorer
Google AI Developers@googleaidevs
We just released Gemini 2.0 Flash Experimental ⚡ Available in the Gemini API and Google AI Studio for testing, it allows developers to build interactive experiences with better performance and multimodal capabilities. goo.gle/3BriaXd
English
Sebastian Riedel (@[email protected]) retweetledi
I am hiring for RS/RE positions! If you are interested in language-flavored multimodal learning, evaluation, or post-training apply here 🦎 boards.greenhouse.io/deepmind/jobs/…
I will also be #NeurIPS2024 so come say hi! (Please email me to find time to chat)
English
Sebastian Riedel (@[email protected]) retweetledi
Sebastian Riedel (@[email protected]) retweetledi
It's paper review day (every day) - since I discovered that DeepMind already knows everything let's look at their latest Paper
arxiv.org/pdf/2411.16679
Lisan al Gaib@scaling01
bro how is it always DeepMind I might just need to go through all DeepMind papers chronologically to build AGI
English
Sebastian Riedel (@[email protected]) retweetledi
🚨 New Paper 🚨
Can LLMs perform latent multi-hop reasoning without exploiting shortcuts? We find the answer is yes – they can recall and compose facts not seen together in training or guessing the answer, but success greatly depends on the type of the bridge entity (80%+ for country, 6% for year)! 1/N
GIF
English
Frontier models can do this stuff, but also not! Opinions differ on how much we even want this (CC @geoffreyirving), but understanding the patterns will be critical regardless.
Been a pleasure to work with Latent Reasoning Dream Team @soheeyang_ @megamor2 @KassnerNora!
Sohee Yang@soheeyang_
🚨 New Paper 🚨 Can LLMs perform latent multi-hop reasoning without exploiting shortcuts? We find the answer is yes – they can recall and compose facts not seen together in training or guessing the answer, but success greatly depends on the type of the bridge entity (80%+ for country, 6% for year)! 1/N
English
@PMinervini Beyond, of course, the @ucl_nlp papers. They were easy to rank for you. (Trying to lean into the conspiracy vibe on this website, how am I doing?)
English
TBH, all papers in my batch were outstanding; it was really tricky to rank them
English
Sebastian Riedel (@[email protected]) retweetledi
The team @jhamrick and I co-lead is hiring a research engineer. If you are interested in improving the capabilities of LLMs in the planning and reasoning space, and building generally capable agents, please apply!
boards.greenhouse.io/deepmind/jobs/…
English
@PMinervini @Niel_Eu25 Oh you mean because in LLMs it depends on the prompt or controller?
Dunno, to me the statement is fine. I would read it as “no matter how I use it, it seems to have problems generating plans.”
English
@riedelcastro @Niel_Eu25 not sure -- it feels a bit like saying "Python can't generate good plans" -- it may depend heavily on how you program it no?
English
Is it me or "LLMs can't plan" sounds like "Python can't plan"? (what does that even mean?)
English
@PMinervini @Niel_Eu25 Are you okay with saying an LLM can (or cannot) generate good plans?
English
@Niel_Eu25 Ok let's replace "Prolog can't do planning" with "A Prolog AI can't do planning" -- does it magically make sense to you now? 🙂
English
Sebastian Riedel (@[email protected]) retweetledi
Excited to share our prompt tuning playbook! (Not an official product. Just authors tips & tricks for better prompting). I'm most excited about first half on mental models for post-training & prompting. Feedback/forks welcome! #LLM #PromptEngineering
github.com/varungodbole/p…
English
@Tim_Dettmers @allen_ai @CarnegieMellon @Titus_vK Oh wow, I haven’t been tracking this website for a while so I missed this at the time. This is fantastic, so happy for you and so well deserved 🤩
English
After 7 months on the job market, I am happy to announce:
- I joined @allen_ai
- Professor at @CarnegieMellon from Fall 2025
- New bitsandbytes maintainer @Titus_vK
My main focus will be to strengthen open-source for real-world problems and bring the best AI to laptops 🧵
English
Sebastian Riedel (@[email protected]) retweetledi
I am hiring for a research engineering role in NYC, focused on Gemini post training. If you are interested, please apply here. Deadline is just in two weeks.
boards.greenhouse.io/deepmind/jobs/…
English
Sebastian Riedel (@[email protected]) retweetledi
I am looking for a Research Scientist intern for 2025. If you have already published work that involves understanding behaviours of AI models looking at their parameters and activations, I would like to hear from you. metacareers.com/jobs/556063310…
English
@howdataworks @egrefen @karlmoritz @_rockt @HeinrichKuttler I bring minimal precision into all my daily routines already. @egrefen is that sufficient?
English
@egrefen @karlmoritz @_rockt @HeinrichKuttler @riedelcastro @egrefen, german aesthetics blend tradition with modern vibes, right? Maybe it’s just about integrating that precision and minimalism into daily routines. What do you think holds folks back?
English
To all the German friends I’ve had the pleasure of working closely with, namely @karlmoritz, @_rockt, @HeinrichKuttler, @riedelcastro, I have one question…
What is stopping you from bringing this particular German aesthetic to everyday life?
youtu.be/D1NdGBldg3w?si…
YouTube
English
Amazing progress @YuxiangJWu and @zhengyaojiang, and great to see the impact of "agent scaffolding" given a base model.
Weco AI@WecoAI
Excited to see OpenAI's recent project, MLE-bench, is based on our open-source effort, AIDE. In their independent evaluation, AIDE surpasses other MLE agents by a large margin!
English
Sebastian Riedel (@[email protected]) retweetledi
We are launching Design Your Own Avatar (DYOA)!
With our latest innovations in multimodal generation at @Creatify_AI , you can now create ultra realistic AI avatars from text description and bring it to life!
This unblocks a whole new level of possibilities.
Check it out: buff.ly/4dsvt6x
Creatify AI@Creatify_AI
THE FUTURE IS HERE! For the first time ever, you can now create your own 100% AI avatar from scratch. Simply describe your ideal representative, from their appearance to their surroundings, and our technology will bring your vision to life! Available now at: buff.ly/4dsvt6x (examples in thread) [No real people were filmed in the making of this video.]
English
Sebastian Riedel (@[email protected]) retweetledi
🚨NEW BENCHMARK🚨
Are LLMs good at linguistic reasoning if we minimize the chance of prior language memorization?
We introduce Linguini🍝, a benchmark for linguistic reasoning in which SOTA models perform below 25%.
w/ @b_alastruey, @artetxem, @costajussamarta et al.
🧵(1/n)
English