Omer Levy retweetledi
State-organized rallies amid anti-regime protests:
Pro-Khamenei rally, Iran, Jan 2026.
Pro-Assad rally, Syria, Nov 2011.
#IranProtests
English
Omer Levy
1.2K posts
Omer Levy
@omerlevy_
AI Researcher at Google DeepMind
Sunnyvale, CA Katılım Mayıs 2014
979 Takip Edilen3.9K Takipçiler
Omer Levy retweetledi
The Palestinian-Israeili conflict is not about Arabs losing land. Since 1948 Arab states lost vast territories in wars and political collapses and no one cared. Sudan lost a third of its territory when South Sudan became independent. Yemen split and collapsed. Morocco lost control over large areas. They adapted. They moved on.
This conflict is not about Muslims losing land. Since 1948 Muslim countries lost territory across Asia and Africa. Pakistan lost East Pakistan which became Bangladesh after a war that killed millions of Muslims. No endless victimhood. No cult of reversal.
This conflict is not about occupation. Since 1948 Turkey occupied Northern Cyprus. Morocco occupied Western Sahara. Syria occupied Lebanon for decades. None turned defeat into an identity.
This conflict is not about refugees. Since 1948 tens of millions were displaced. They rebuilt countries. They rebuilt lives. We were deliberately taught not to rebuild anything.
This conflict is about choice.
A choice to reject every compromise. A choice to glorify loss instead of fixing it.
A choice to raise children on slogans instead of skills.
A choice to turn defeat into religion and failure into virtue.
No one is trapped. This is self-inflicted.
And it continues because admitting defeat would require accountability.
English
Omer Levy retweetledi
glad everyone is enjoying our model, we have no plans on slowing down either 😊
English
Omer Levy retweetledi
Gemini 3 Pro sets new record on SWE-bench verified: 74%! (evaluated with minimal agent) Costs are 1.6x of GPT-5, but still cheaper than Sonnet 4.5. Gemini iterates longer than everyone; run your agent with a step limit of >100 for max performance. Details & full agent logs in 🧵
English
Omer Levy retweetledi
Excited to share that a scaled up version of Gemini DeepThink achieves gold-medal standard at the International Mathematical Olympiad. This result is official, and certified by the IMO organizers.
Watch out this space, more to come soon!
deepmind.google/discover/blog/…
English
Omer Levy retweetledi
Very excited to share that an advanced version of Gemini Deep Think is the first to have achieved gold-medal level in the International Mathematical Olympiad! 🏆, solving five out of six problems perfectly, as verified by the IMO organizers! It’s been a wild run to lead this effort and I am grateful to everyone in the team for such an amazing achievement! Blog post in the thread and more to share soon!
Thang Luong@lmthang
Super thrilled to share that our AI has has now reached silver medalist level in Math at #imo2024 (1 point away from 🥇)! Since Jan, we now not only have a much stronger version of #AlphaGeometry, but also an entirely new system called #AlphaProof, capable of solving many more Olympiad problems. This is a large-scale project that I was fortunate to co-lead at @GoogleDeepMind! See our blog & NYT articel below! Blog: dpmd.ai/imo-silver NYT: nytimes.com/2024/07/25/sci…
English
Omer Levy retweetledi
Flying to Singapore 🇸🇬 for @iclr_conf this week!
Looking forward to catching up with friends and discuss multimodal modeling or our new π-0.5 physicalintelligence.company/blog/pi05 (and other Visual language action models).
English
Omer Levy retweetledi
pretty mind-blowing fact I just learned about transformer language models:
the positional embeddings don't really do anything. you can just get rid of them and the model still works just as well
sounds impossible, doesn't it?
turns out standard LLMs aren't actually permutation-invariant because of the causal mask. so they just learn somehow to "figure out" what position they're at by counting the number of tokens they can see at a given position
p crazy
English
Omer Levy retweetledi
Transformers can work without using positional embeddings at all. Llama 4 uses positional embs for local attn but not globally. Our paper from 2022 shows why this works- the causal mask allows transformers to infer positions. arxiv.org/pdf/2203.16634
English
Omer Levy retweetledi
New paper alert!
Curious how small prompt tweaks impact LLM accuracy but don’t want to run endless inferences? We got you. Meet DOVE - a dataset built to uncover these sensitivities.
Use DOVE for your analysis or contribute samples -we're growing and welcome you aboard!
Eliya Habba@EliyaHabba
Care about LLM evaluation? 🤖 🤔 We bring you🕊️ DOVE a massive (250M!) collection of LLMs outputs On different prompts, domains, tokens, models... Join our community effort to expand it with YOUR model predictions & become a co-author!
English
Omer Levy retweetledi
Omer Levy retweetledi
DeepMind has ambitious plans to make massive generative models that simulate the world. I'm hiring for a new team with this mission. Come build with us!
boards.greenhouse.io/deepmind/jobs/… boards.greenhouse.io/deepmind/jobs/…
English
Omer Levy retweetledi
Introducing Gemini 2.0 Flash Thinking, an experimental model that explicitly shows its thoughts.
Built on 2.0 Flash’s speed and performance, this model is trained to use thoughts to strengthen its reasoning.
And we see promising results when we increase inference time computation!
English
Omer Levy retweetledi
We’ve been *thinking* about how to improve model reasoning and explainability
Introducing Gemini 2.0 Flash Thinking, an experimental model trained to think out loud, leading to stronger reasoning performance.
Excited to get this first model into the hands of developers to try out!
English
Omer Levy retweetledi
Everyone’s talking about synthetic data generation — but what’s the recipe for scaling it without model collapse? 🤔
Meet ALMA: Alignment with Minimal Annotation.
We've developed a new technique for generating synthetic data and aligning LLMs that achieves performance close to Llama3-instruct with only 9000 labeled examples, that's less than 1% of the millions of human annotation data typically used for alignment.
Check out the full paper here: arxiv.org/abs/2412.04305
English
Omer Levy retweetledi
(1/5) 🎉[New Paper] Altogether: Image Captioning via Re-aligning Alt-text arxiv.org/abs/2410.17251 is accepted by #EMNLP2024 : we re-align existing alt-texts as captions for images instead of captioning images from scratch (e.g. MS-COCO captioning).
English
Omer Levy retweetledi
Excited to share our latest work on multimodal language modeling! We introduce Mixture-of-Transformers (MoT), a sparse architecture that jointly trains modality-specific transformers for text, image, and speech data. Each transformer exclusively processes its own modality's tokens while attending to others, leading to enhanced training efficiency and flexibility. (arxiv.org/abs/2411.04996)
Highlights:
🚀 2-3x more FLOP-efficient compared to dense models across different settings.
🧠 Applied to different architectures (Chameleon and Transfusion) and modalities (text, image, speech).
⚙️ Enables flexible scaling of transformer towers per modality (e.g., extend text transformer to MOE; more parameters for text).
📈 Performance improvements are maintained after fine-tuning.
Checkout @weixin's post for more information.
Weixin Liang@liang_weixin
How can we reduce pretraining costs for multi-modal models without sacrificing quality? We study this Q in our new work: arxiv.org/abs/2411.04996 At @AIatMeta, We introduce Mixture-of-Transformers (MoT), a sparse architecture with modality-aware sparsity for every non-embedding transformer parameter (e.g., feed-forward networks, attention matrices, and layer normalization). MoT achieves dense-level performance with up to 66% fewer FLOPs! ✅ Chameleon setting (text + image generation): Our 7B MoT matches dense baseline quality using just 55.8% of the FLOPs. ✅ Extended to speech as a third modality, MoT achieves dense-level speech quality with only 37.2% of the FLOPs. ✅ Transfusion setting (text autoregressive + image diffusion): MoT matches dense model quality using one-third of the FLOPs. ✅ System profiling shows MoT achieves dense-level image quality in 47% and text quality in 75.6% of the wall-clock time** Takeaway: Modality-aware sparsity in MoT offers a scalable path to efficient, multi-modal AI with reduced pretraining costs. Work of a great team with @liliyu_lili, Liang Luo, @sriniiyer88, Ning Dong, @violet_zct, @gargighosh, @ml_perception, @scottyih, @LukeZettlemoyer, @VictoriaLinML.👏 **Measured on AWS p4de.24xlarge instances with NVIDIA A100 GPUs.
English
Omer Levy retweetledi
I’m so disgusted by what I just witnessed. In Amsterdam, Israelis & Jews leaving a soccer match were beaten unconscious by mobs, thrown in the river, and forced to say “free Palestine.” This is the direct result of normalizing antisemitism post Oct. 7, where the most flagrant acts of Jew-hatred were cast aside as just “bleeding hearts who are against the war.”
Remind me again how these are all just kind people who care about humanity.
If you can’t bring yourself to speak up now—after failing the Jewish community for the past year with silence and/or by insisting that “globalize the intifada” was anything other than a dogwhistle to harass Jews around the world—you have no right to call yourself a humanitarian.
English