Oriol Vinyals

@OriolVinyalsML223k followers

Co-founder & CTO of Discovery Loop; Partner, 224 Ventures Past: Google DeepMind VP, Gemini TL (1.0 to 4.0), Alpha{Fold,Star,Code}, seq2seq, distillation

Open on X

Top outlier

18.5×

Best organic post

Peak views

715k

Single post

Bookmarks

11k

Across 97 organic posts

Last scan

Oct 17

Community cache

Posts 97 posts

#	Date	Post	↓ Outlier	Views	Likes	Reposts	Bookmarks
1	Oct 2, '22	This neural network architecture that was showcased at the @Tesla AI day is a perfect example of Deep Learning at its finest. Mix and match all the greatest innovations to do somet	18.5×	0	5.4k	534	448
2	May 17, '24	Today we have published our updated Gemini 1.5 Model Technical Report. As @JeffDean highlights, we have made significant progress in Gemini 1.5 Pro across all key benchmarks; TL;DR	17.3×	715k	976	188	314
3	Aug 5	Thirteen years ago, a quick email exchange with @JeffDean brought me to @Google. After an unforgettable journey from Brain, to DeepMind, and to @GoogleDeepMind, working on seq2seq	12.4×	512k	4.2k	193	408
4	Mar 24, '23	To whoever invented USB-C: THANK YOU!!	11.8×	490k	732	59	6
5	Dec 6, '23	Exciting times, welcome Gemini (and MMLU>90)! State-of-the-art on 30 out of 32 benchmarks across text, coding, audio, images, and video, with a single model 🤯 Co-leading Gemini ha	11×	457k	1.9k	261	318
6	Jun 17, '25	Hello Gemini 2.5 Flash-Lite! So fast, it codes each screen on the fly (Neural OS concept 👇). The frontier isn't always about large models and beating benchmarks. In this case, a	10.2×	423k	1.6k	255	749
7	Feb 15, '24	Gemini 1.5 has arrived. Pro 1.5 with 1M tokens available as an experimental feature via AI Studio and Vertex AI in private preview. Then there’s this: In our research, we tested Ge	9.2×	381k	862	162	116
8	Nov 30, '20	Biology's ImageNet moment has arrived: #AlphaFold2 solves Protein Folding at the #CASP14 competition! Blog:	9.0×	0	2.2k	442	125
9	Jan 3, '18	"Optimal" running, brought to you by Deep RL.	7.9×	0	1.8k	578	3
10	Jan 11, '23	Many people mistakenly think that large language models that generate one word at a time is the end game. Connecting LLMs with tools, e.g.: search engines, python interpreters, etc	7.5×	310k	1.3k	182	456
11	Jan 22, '25	I uploaded my PhD thesis (98K tokens) and asked our latest Flash thinking model to find flaws in it. "Potential over-reliance on empirical results without sufficient theoretical gr	7.2×	300k	603	40	139
12	Jan 1, '20	The "Deep Learning Toolbox" has greatly expanded in the last decade thanks to our wonderful research community. Also, important progress has been made to make our community more in	6.4×	0	1.3k	349	302
13	Aug 5, '25	Incredible evolution of "Neural Video Games": from GQN (2018) to Genie3 (2025). The future is exciting!	6.3×	260k	2.9k	319	755
14	Sep 29, '18	Best GAN samples ever yet? Very impressive ICLR submission! BigGAN improves Inception Scores by >100. Paper: Lots more samples:	6.3×	0	1.3k	487	80
15	Feb 6, '25	Distillation has been on the news (!) due to @deepseek_ai. The paper was actually rejected from NeurIPS 2014 due to lack of novelty 🧐 (true-ish), and lack of impact 🙃. Thanks rev	6.2×	258k	1.5k	163	628
16	May 6, '25	Ahead of I/O, we’re releasing an updated Gemini 2.5 Pro! It’s now #1 on WebDevArena leaderboard, breaking the 1400 ELO barrier! 🥇 Our most advanced coding model yet, with stronger	6.1×	254k	770	67	104
17	Nov 15, '25	🤔🤔🤔	6.1×	253k	530	9	41
18	Dec 7, '23	seq💜💙❤️seq @ilyasut @OriolVinyalsML @quocleix	5.9×	244k	1.2k	59	111
19	Jan 24, '19	Happy that we could share #AlphaStar progress with you all! Good Games @LiquidTLO and @Liquid_MaNa, and @Artosis and @RotterdaM08 for a great show! You can see all the details in t	5.7×	0	1.3k	421	27
20	Nov 5, '21	The Deep Learning Devil is in the Details. I love this work from @IrwanBello and collaborators in which they show how training "tricks" improve ~3% absolute accuracy on ImageNet, p	5.6×	0	1.2k	220	332
21	Apr 20, '23	𝗚𝗼𝗼𝗴𝗹𝗲 𝗗𝗲𝗲𝗽𝗠𝗶𝗻𝗱	5.5×	228k	800	82	15
22	Feb 15, '24	To show what’s possible with the drastically huge context window in Gemini 1.5 Pro, we prompted it with the three.js examples code - over 100,000 lines of code/800k+ tokens! (That’	5.3×	219k	954	176	235
23	Jun 26, '18	Welcome back, gradients! This method is orders of magnitude faster than state-of-the-art non-differentiable techniques. DARTS: Differentiable Architecture Search by Hanxiao Liu, Ka	5.2×	0	1.2k	382	77
24	Mar 25, '25	Introducing Gemini 2.5 Pro Experimental! 🎉 Our newest Gemini model has stellar performance across math and science benchmarks. It’s an incredible model for coding and complex reas	5.1×	213k	1.1k	153	204
25	Feb 29, '20	Transformers are a special case of Graph Neural Networks. This may be obvious to some, but the following blog post does a good job at explaining these important concepts.	4.7×	0	936	232	257
26	Dec 7, '23	Really happy to see the interest around our “Hands-on with Gemini” video. In our developer blog yesterday, we broke down how Gemini was used to create it. We gave Gemini sequences	4.4×	181k	562	108	230
27	Jul 21, '25	Drastic progress on maths with Gemini 2.5! As a math undergrad, I am impressed 🤯 🥈 -> 🥇 ✅ Formal -> Informal ✅ Specialized model -> General model ✅ Available soon ✅ Hug	4.4×	181k	747	76	81
28	Nov 18, '18	Very interesting visualization of 80 years of #AI research represented as connectionist (~neural nets) vs symbolic (~rule based) . I wonder what this will look like in another 80 y	4.3×	0	841	383	62
29	Jul 10, '17	Agents having fun in Parkour! Cool paper from colleagues at DeepMind	4.2×	0	798	411	0
30	Jun 7, '23	It took a lot of Deep RL finesse (& some Transformers ❤️) to get the first change in over a decade accepted into the LLVM standard C++ library. AlphaDev, published @Nature by m	4.2×	172k	629	78	48
31	Jan 27, '24	Fun game. Clocking 17973 citations: "Distilling the knowledge in a neural network" @geoffreyhinton, @OriolVinyalsML, @JeffDean Reviewer 38 (NeurIPS 2014): "This work is incremental	4.1×	171k	449	43	81
32	Oct 3, '20	Recent conversation with a friend: @ilyasut: what's your take on ? @OriolVinyalsML: my take is: farewell convolutions : )	4.1×	0	840	178	212
33	Mar 21, '24	We are rolling out Gemini 1.5 Pro API so that you can keep building amazing stuff on top of the model like we've seen in the past few weeks. Also, if you just want to play with Gem	4.0×	166k	669	124	227
34	Feb 20, '24	Gemini 1.5 Pro launched last week and already we're seeing the community produce some amazing interactions with long context. Below 👇 are some highlights and cool posts from folks	3.9×	161k	424	74	133
35	Feb 11, '22	Google Maps 2.0 aka Neural Google Maps aka NERF Google Maps. Amazing!	3.8×	0	1k	192	82
36	Feb 4, '23	@ylecun Remind me in 5 years... 🍿	3.8×	156k	530	9	7
37	Jun 12, '18	Graph Neural Networks / Relational Networks are models worth studying. We wrote a pretty comprehensive review about them which I hope you will find helpful (code forthcoming!).	3.7×	0	808	260	76
38	Nov 27, '24	Such a great honor, thanks a lot @NeurIPSConf and congrats to my esteemed co-authors @ilyasut & @quocleix! The 2014 talk also stood the test of time IMO. Here is a slide from it (p	3.5×	145k	673	72	139
39	Oct 4, '17	WaveNet becomes the voice of Google Assistant! 1000x faster & better, brought to you by a neural net : )	3.5×	0	731	311	0
40	Mar 12, '25	Gemini 2.0 Flash debuts native image gen! Create contextually relevant images, edit conversationally, and generate long text in images. All totally optimized for chat iteration. Tr	⚡ 3.4×	846k	636	83	224
41	Feb 2, '22	#AlphaCode: a drastic version of seq2seq & search. The Bitter Lesson strikes again, exposing the power of general methods that scale w/ increased compute! Like AlphaStar, an ag	3.3×	0	853	132	97
42	Feb 14, '20	Rapid unsupervised learning progress thanks to contrastive losses, approaching supervised learning! -40% Multitask SSL (2017) -50% CPC (2018) -70% AMDIM/MOCO/CPCv2/etc (2019) -76.5	3.1×	0	622	193	125
43	Mar 16, '25	Mind blown 🤯 Gemini 2.0 Flash interleaved image generation is a game changer. I explained AlphaFold to my 2yo daughter's class for science week on Friday 🧑‍🔬. Prompt: "Write an	3.0×	124k	639	59	207
44	Jul 23, '21	#AlphaFold is Deep Learning at its finest. Take a (small!) dataset of aminoacid sequences -> 3D structure. Use some of the best ideas from ML, NLP, CV (transformers, GNNs, BERT,	2.9×	0	712	147	87
45	Jul 26, '22	It's out! Thanks @lexfridman for the interesting conversation! I will make sure to acknowledge you in some future papers as our chat already sparked interesting research ideas!	2.9×	0	902	74	50
46	Jul 23, '25	We already are. Just through a fairly narrow communication bottleneck.	2.8×	116k	673	21	67
47	Apr 22, '20	Datasets over Algorithms. Creating good benchmarks and datasets is critical for the advancement of #AI. Thanks ImageNet, WSJ, ALE, PTB, GLUE, WMT, etc.! (Old) Source:	2.8×	0	647	172	63
48	Dec 10, '23	On my way to @NeurIPSConf #NeurIPS2023 where I'm presenting 0 papers 🙃 Come chat with the Gemini ♊️ team at @GoogleDeepMind / @GoogleResearch booths on Tue/Wed from 1:30 to 3:00 P	2.7×	111k	344	40	42
49	Sep 23, '19	Rapid Learning or Feature Reuse? Meta-learning algorithms on standard benchmarks have much more feature reuse than rapid learning! This also gives us a way to simplify MAML -- (Alm	2.7×	0	579	154	90
50	Jun 4, '19	Surprising how simple ideas can yield such a good generative model! -Mean Squared Error loss on pixels -Non-autoregressive image decoder -Discrete latents w/ straight through estim	2.7×	0	597	157	78
51	Sep 30, '20	Internships are a great learning opportunity. I learned a lot during my 5 (!) internships, and kept learning as well as an intern host. Things I learned as a host? VAEs, GANs, GNNs	2.6×	0	589	111	110
52	Aug 9, '17	Launching StarCraft2 Learning Environment (Linux)! A3C baseline on mini & full game + policies trained on 1M replays.	2.6×	0	553	241	2
53	Jan 22, '23	Zero accepted papers to #ICLR2023 😢 / Zero submitted 🙃. Acceptance rate = NaN #achievementunlocked	2.6×	106k	347	9	4
54	Jan 15, '20	Although neural networks usually require massive datasets to do impressive things, for me the highlight of #AlphaFold is the fact that it achieved state-of-the-art using only 30K t	2.5×	0	597	150	62
55	Mar 27, '18	RL + GANs: Program synthesis with an agent that uses a paint program to fool a discriminator. Paper+Blog:	2.4×	0	537	191	27
56	Aug 2, '22	Interested in exploring the limits of large-scale models? Consider joining our brand new Scalable Deep Learning team in California as a Research Engineer or Research Scientist, see	2.3×	0	525	59	143
57	Aug 25, '19	A good example of our field moving (too) fast: the all attention layer was actually released in the original transformer paper from 2017, with similar findings. Code: Paper: (only	2.3×	0	502	137	77
58	Feb 10, '19	Most surprising facts about #AlphaStar: 1. A Neural Network can play full StarCraft -- without hard coding 2. Pure imitation learning achieves Gold/Platinum 3. Naive Self-Play "get	2.3×	0	559	125	33
59	Aug 23, '16	Excited to have won the MIT TR35 innovator award! Very grateful to do research alongside brilliant colleagues : )	2.2×	0	557	122	10
60	Jul 10, '19	Excited that AlphaStar will be joining for scientific evaluation & for a limited time! The updated agent: -Plays as and versus all races -Plays multiple ladder maps -Uses a cam	2.1×	0	539	138	9
61	May 22, '25	Yesterday at #GoogleIO, we introduced Gemini 2.5 Pro Deep Think 🧠, pushing the frontiers of AI reasoning. This enhanced reasoning mode is built to tackle drastically complex probl	2.1×	87k	619	61	106
62	Sep 8, '21	If you love Graph Neural Nets, language & Wikipedia, this is for you! We open sourced data & baselines for the challenging task of generating text from knowledge graphs (&a	2.1×	0	443	88	116
63	Nov 14, '20	One of the first things I did after joining @Google was to implement an LSTM using TF's predecessor "DistBelief". The model definition had 16245 lines! Once we moved to @TensorFlow	2.1×	0	719	36	28
64	Feb 25, '21	2016 -> 2021 Understanding Deep Learning (Still) Requires Rethinking Generalization A "refreshed" version of our paper from 2016, with a cool perspective from @prfsanjeevarora @	2.0×	0	419	83	122
65	Apr 12, '25	We are blown away by the reception of Gemini 2.5 Pro, and the many positive surprises from benchmarks and community evaluations. That is the ultimate leaderboard!	2.0×	84k	529	29	41
66	May 17, '19	The distillation paper was actually rejected from #NeurIPS2014 : ) As I'm sure many people are working towards #NeurIPS2019, optimize for long term impact, not probability of accep	2.0×	0	560	93	36
67	Apr 6, '21	I am proud to have been the very first report of Samy Bengio. His kindness, values, and leadership, is what I pursue as a manager. Thank you, Samy!	2.0×	0	625	66	22
68	May 20, '25	Today we introduced Gemini Diffusion⚡️ (& DeepThink, Veo3, Imagen4, 2.5 updates...). It's been a dream of mine to remove the need for "left to right" text generation. It's so fast,	1.9×	80k	761	95	137
69	May 8, '18	Very impressive application of deep learning from Chen Chen & @intel for "Seeing in the Dark" with regular cameras. Code available, too!	1.9×	0	415	157	21
70	Jan 23, '19	We’ve spent the past several months hard at work on #StarCraft2 - tomorrow we’ll be hosting a livestream demonstration! Excited to hear what you all make of it : ) Tune in @ 6:00pm	1.9×	0	491	100	7
71	Jun 28, '19	Great post by Prof. David McAllester on why discrete representations matter, based on our findings in VQ-VAE2. "Vector quantization seems to be a minimal-bias way for symbols to en	1.8×	0	366	111	72
72	Feb 6, '18	Evolution > RL (for now...) for architecture search. New SOTA on CIFAR10 (2.13% top 1) and ImageNet (3.8% top 5). 🔥 450 GPU / 7 days & 900 TPU / 5 days 🔥	1.8×	0	394	153	5
73	Apr 29, '21	Linear Algebra and Calculus is really all you need.	1.8×	0	473	43	73
74	Jul 12, '18	Universal Transformers propose to augment Transformers with Recurrence in depth and Adaptive Computation Time. This model outperforms Vanilla Transformers in MT / bAbI / LA / LTE.	1.7×	0	366	131	40
75	Nov 14, '24	What wall?	1.7×	72k	472	29	30
76	Sep 10, '20	Interested in a PhD in ML? Applications are open to anyone wishing to pursue a PhD in ML within ELLIS. Excited to be working with many talented students! Portal opens: Oct 1st De	1.7×	0	368	83	73
77	Dec 8, '21	This is a great thread, +1e100. @karpathy didn't mention my (biased!) favorite example of 2021. Models designed to generate words (transformers) & model language (BERT) were re	1.7×	0	443	52	78
78	Jan 4, '19	"Failing" is too often portrayed as something negative. Failures, and not successes, have been a higher drive for me both personally and professionally (and there have been plenty!	1.6×	0	457	85	5
79	Jan 7, '21	Neural networks still surprise me as rich information retrieval systems. The generalization and "creativity" exhibited by @OpenAI's DALL·E is amazing! Good luck finding these image	1.6×	0	459	67	24
80	Nov 12, '21	The intersection of AI + Art is fascinating! Our take on this is presented in Arnheim 2. As an example of what you could do with it, a drawing of @ylecun using only MNIST digits. #	1.6×	0	399	58	49
81	Oct 30, '19	#AlphaStar @nature: Grandmaster level as all 3 @StarCraft races on , w/ a pro approved interface (camera & APM limits). 2 years ago I thought this was impossible! How? Imitatio	1.6×	0	393	95	7
82	Apr 10, '18	Great SIGGRAPH paper which uses RL/ML for motion generation. I predict many more applications of machine learning to graphics & other domains -- looking forward to that!	1.6×	0	353	130	11
83	Nov 27, '17	Great stuff from @maxjaderberg et al on hyperparameter search based on evolving populations that actually outperforms random search!	1.5×	0	336	125	0
84	Nov 22, '17	Almost 3000x speedup with Parallel WaveNet (Google Speech production model). Brought to you by @avdnoord et al. Drastic! More details & link to paper:	1.4×	0	355	111	0
85	Jun 21, '25	@elonmusk If you delete all errors, the model may lose the ability to detect and correct errors 🤔	1.4×	59k	343	8	22
86	Jul 19, '17	The LSTM Strikes Back: sub 60 perplexities on LM PTB (single model). All you need is a (well tuned) LSTM :)	1.4×	0	347	104	4
87	Nov 7, '20	May our gradients never vanish. #sciencewins	1.4×	0	474	46	3
88	May 12, '23	"Context length" is the new "model size" 🤔	1.4×	57k	282	20	17
89	Oct 2, '22	With all the debates on who-claimed-what-when, ego wars, research-prophecies, and other non-sense that is filling my Twitter feed these days, it is sobering to see what good execut	1.4×	0	500	22	3
90	May 25, '22	Summary of the last month (!) of projects released in Deep Learning. Huge congrats to the teams involved! #Imagen #flamingo #dalle2	1.3×	0	366	45	36
91	Jul 25, '24	AI is beating me at the things I love, one step at a time (coding, StarCraft, mathematics, ...). AlphaProof solved the most difficult 2024 IMO problem (P6). Answer doesn't fit in t	1.3×	53k	423	39	65
92	May 7, '23	Best feeling in the (Neural Network) world: loss is going down 📉 Happy Sunday!🌞	1.2×	50k	167	8	3
93	Jun 24, '22	Had a great time chatting with @lexfridman earlier today, looking forward to the episode!!	1.1×	0	387	16	11
94	Dec 14, '24	seq2seq 10 years later. 💙@ilyasut @quocleix. Talk 👇	1.1×	46k	365	31	58
95	Mar 26, '25	A mere ~16 point jump on . Such a good model! Gemini 2.5 Pro ♊️	1.1×	46k	345	27	23
96	Dec 6, '24	A year ago we launched Gemini 🚀 Back then I shared with the team that it would be great to be in a position to beat ourselves. Just a year later this is starting to happen. But we	1.0×	42k	350	53	26
97	Dec 11, '24	Gemini 2.0 Flash ⚡️ has arrived! 2.0 Flash > 1.5 Pro (again!) 📈 Interacts with a browser 🤖 Native image generation 🖼️ and much more! Try it out As a preview of what is possib	1.0×	41k	349	58	72
98	May 28, '21	Few Shot Learning Is All You Need #vaccinated #NeurIPS2021	1.0×	0	369	14	1
99	Jul 20, '22	Great to be at #ICML2022! I missed poster sessions in person. Deep Learning Is All I Need ♥️	1.0×	0	385	4	3
100	Oct 9, '24	Richly deserved @NobelPrize prize for @geoffreyhinton, @demishassabis, and John Jumper. I've had the privilege to work alongside these brilliant minds. Here are some personal insig	0.8×	34k	430	36	53
101	Aug 26, '25	We've just upgraded Gemini 2.5 Flash image generation & editing! 🍌🍌🍌 Besides topping leaderboards, it topped my model usage this month. It keeps subjects consistent, you can	0.7×	29k	393	20	21

Scan another creator

Run this on any account.

XOutlier builds this table for every public profile you open on x.com, with live data and every post, not just the top 50.

Add to Chrome Download .zip