XOutlier
Download .zipAdd to Chrome

Teknium 🪽

@Teknium134k followers

Cofounder and Lead Engineer - Hermes Agent @NousResearch, prev @StabilityAI Github: HuggingFace:

Open on X
Top outlier
134×
Best organic post
Peak views
3.4M
Single post
Bookmarks
40k
Across 91 organic posts
Last scan
Oct 16
Community cache
Posts 91 posts
#DatePost↓ OutlierViewsLikesRepostsBookmarks
1Aug 7, '25Umm what is this new chart crime?134×3.4M4.4k191368
2Jan 30, '25This is the entire code needed to reproduce R1 lol Hundreds of Billions of Dollars Later92.7×2.3M18k1.5k13k
3Aug 9, '25OpenAI has done some real damage81.4×2.1M4.6k125726
4Feb 18, '25In my testing it was at least as good in thinking mode then o3-full deep research was, despite that not being listed here - Interesting to note that grok-3mini seems generally bett67.5×1.7M3.5k410202
5Feb 25We just released Hermes Agent! In my humble opinion a very good blend between coding agents like Claude Code and generalist agents like Clawdbot. Been working on this for the last 43.7×1.1M2.4k141861
6Jan 26, '25Its crazy deepseek direct api has seemingly no rate limits of any kind30.3×768k7.2k2531.1k
7Apr 19, '24Welp folks, we have gpt-4 at home30.2×765k4.1k335822
8May 3, '25I think web search is easy but how are they getting past all the captchas and such?26.6×673k1.7k32514
9Aug 7, '25Amazing..23.4×591k5.8k155318
10Apr 18, '24Holy shit lol21.9×554k1k69172
11Feb 27, '25I really need an LLM that reads in my actual whole codebase and lets me QA it. Cursor afaict doesn't do this. What does?20.2×512k1.8k651.3k
12Jul 12, '25Literally never trust this guy - as far as i can tell, all the big providers already have the model IN HAND, and were ready to start serving it. A 1T parameter model that matches o17.9×452k3k162454
13Jan 4, '25lol this exchange was funny to me16.9×428k16k443752
14Oct 6, '25I’m confused - OpenAI is buying chips so amd gave them 10% of their company?16.9×426k2.3k54265
15Oct 7, '23Introducing Mistral Trismegistus 7B - the first instruction dataset on the Esoteric, Spiritual, Occult, Wisdom Traditions, and all things paranormal trained on Mistral, and possibl15.3×388k1k136793
16Mar 24, '25A wild Deepseek has appeared15.2×385k3.4k300869
17Jan 1, '24Happy New Years Everybody! 🥳 One year ago today, I had: - Never trained any model - Did not know the first thing about AI - Never worked in Tech - Had 8 followers on twitter? (pro14.5×366k2.1k102631
18Jan 15, '24It's finally time! Our Mixtral 8x7B model is up and available now! Nous-Hermes-2 Mixtral 8x7B comes in two variants, an SFT+DPO and SFT-Only, so you can try and see which works bes12.9×325k1.5k204541
19Aug 5, '25Starting to feel like this gpt oss was trained on like 20T tokens of distilled safe maybe even benchmaxxed data from o3. There seems to be no base model underneath.. Is this phi 5 12.6×319k1.7k57416
20Mar 16, '23Incredible. I gave GPT-4 this insanely complex image, and it worked.12.3×311k927118251
21May 28, '25Deepseek r1-v2 is out 👀11.9×302k3.1k331694
22Mar 29, '254o Imagen can do calcuations *during* it's image generation somehow11.7×296k2.5k61256
23Apr 19, '24I think Meta and Llama-3 is the final nail in the coffin to several misconceptions I've been fighting against for the last year. Llama-3 Chat was trained on over 10M Instruction/Ch11.3×285k1.1k96486
24Sep 8, '25New challenge now that models are overfit on the original lol10.9×276k5.1k104205
25Oct 6, '25You can now invest in Nvidia, Intel, AMD, ARM, OpenAI, Mistral, CoreWeave, Nebius, and more with just one ticker: NVDA lol10.9×275k5k210426
26Jun 3, '23Announcing Nous-Hermes-13b - a Llama 13b model fine tuned on over 300,000 instructions! This is the best fine tuned 13b model I've seen to date, and I would even argue rivals GPT 310.4×264k1k155597
27Aug 15, '23Why do almost no papers release code, datasets, info on replication, final models, or any combination of these? I thought for science to work results had to be reproducible and ver9.9×251k1.5k11176
28Apr 18, '24These numbers are insane. I can't even imagine what the larger one(s) will be. Looks like Mistral 7B might be dead as of today though, and maybe even sonnet lol My favorite is the 9.8×249k83066124
29Sep 26, '24Yann constantly says Europe is more free then America then puts out a model that Europeans aren't allowed to download because they aren't free enough to do so lol9.4×238k2.8k11985
30Jan 31, '24Today I have a huge announcement. The dataset used to create Open Hermes 2.5 and Nous-Hermes 2 is now PUBLIC! Available Here: This dataset was the culmination of all my work on cur9.4×237k1.9k284818
31Dec 22, '24I havent seen any Sora videos on twitter since launch day..9.3×234k1.9k1284
32Mar 17, '24Grok is out. 320~B Params - 8x33B MoE blog: code: Download: magnet:?xt=urn:btih:5f96d43576e3d386c9ba65b883210a393b68210e&tr=https%3A%2F%%2Fannounce.php%3Fpasskey%3Decac4c57591b8.8×223k950121299
33Jan 30, '25lol Sometimes I think that @Microsoft secretly wants OpenAI to die - they are literally serving R1 for FREE xD8.8×223k3.1k141720
34Nov 2, '23Today I am releasing Open Hermes 2.5! This model used the Hermes 2 dataset, with an added ~100k examples of Code Instructions, created by @GlaiveAI! This model was originally meant8.6×217k1k130419
35Dec 3, '23Announcing Nous Hermes 2.5 Vision! @NousResearch's latest release builds on my Hermes 2.5 model, adding powerful new vision capabilities thanks to @stablequan! Download: Prompt the8.5×216k948162399
36Oct 29, '25why all oai peeps acting like this is some traumatizing hardship openai went through and survived lol7.8×197k1.4k1684
37Jan 20, '25OpenAI seething so hard I cant even paste r1's paper into o1 without a content violation what losers7.5×190k2.9k134286
38Dec 3, '24I want to build a really cool home library like 200-300 books on a series of bookshelves Give me your like, 3 absolute top books, fiction or non-fiction, doesn't matter.7.5×190k880271.1k
39Aug 7, '25Damn lol7.3×185k2.2k67195
40Jul 22, '24If this image is real comparing Llama-3.1 405/70/8b against gpt4o - we have SOTA Frontier Models available Open Source now:7.3×184k1k126197
41Jul 13, '25Had no idea these existed6.9×174k90319192
42Feb 1, '25Let me tell you all a little story. Sometime around a or so year ago i reached out to an openai staffer who will not be named who had implied they would be very interested in doing6.6×167k1.3k144299
43Jan 20, '25Destroys 4o; we have frontier models @ home I guess now6.5×165k2k119562
44May 22, '25I think what needs to be stated about the Claude 4 narc drama is that this is not a just emergent thing because the model's smart, but a direct result of the obssessive and intenti6.5×163k1.2k69204
45Dec 20, '24If its o2/3 Im going to be severely dissapointed. All we want is gpt4.5/5 ffs lol6.3×160k853561
46Feb 28, '25I think this is the most insane part of 4.5 release to me. The knowledge cutoff is 2023 still. How do you even have a current pretraining run that didnt see data past 2023? So many6.3×159k89325115
47Feb 18, '25Grok3 Unveiled6.2×157k1.1k3285
48Feb 16, '25Dude o3-mini-high + deep research when I asked to make code that uses OAI Library + o3-mini-high, gave me a deprecated gpt3.5 api call and used a deprecated API schema (completions6.2×156k1k37141
50Feb 16, '25I almost fell for o3-mini being good this is absurd6.0×151k9111965
51Mar 23, '25Okay so I finally got full @cursor_ai and the agent mode is like Devin but actually works. Crazy that Devin charges $500 a month and this costs 20.. lol5.8×148k1.1k22214
52Jun 8, '25Its funny that Anthropic and Google are the only competitors in coding AI that matter right now openai has just become a gen-media company and I only see people meaningfully using 5.8×147k1.5k53204
53Mar 15, '24This explains why Yann is so bearish on LLMs... 😲5.5×139k94834164
54Feb 27, '25Guys its been 2+ years and 1000s' of times more capital has been deployed since gpt4.. what the hell happened5.4×137k90228120
55Jun 30, '25tfw you know a paper's going to be good5.4×136k1.5k120589
56Dec 22, '24Ok tried Devin on two new tasks, one was automatically creating data visualizations of benchmarks, which took about 4 hours and a ton of back and forth debugging with it to get it 5.3×135k87635169
57Apr 2, '23I have a true gift for LLM Devs and the opensource AI community. Several GPT-4 Generated datasets. Toolformer, Instruct, Roleplay-Instruct, and soon, Code-Instruct datasets, all ge5.3×133k874141605
59Feb 4, '25Really crazy no ones talking about KimiK paper - Its even deeper and more insightful than r1 on the RL measures they went through and produced what they claim is an o1 level multim5.2×132k1.6k1651.1k
61Jan 24, '25Unbelievable the amount of cope, seethe, and hoop jumping people are doing to discredit deepseeks accomplishments lol5.2×131k3k185196
62May 1, '25They couldnt just.. give it to everyone right now? Lol5.1×130k1.8k4859
63Dec 21, '24So is anthropic going to answer openai5.0×127k1.1k2759
64Jun 20, '24Announcing Hermes 2 Theta 70B!! Our most powerful model ever released, and our first model to catch up to GPT4 on MT Bench, and beat llama-3 70B instruct nearly across the board! W4.9×124k983116233
65Sep 30, '23Thanks DALLE3, everyone needed to know what pikachu's musculoskeletal system looked like4.8×122k8516465
67Feb 18, '25Please god votee for o3-mini lmao what ya'll doin we can distill it into a phone sized model people!!4.7×119k1.3k5750
68Feb 18, '25OpenAI is so confusing4.6×118k1.1k2069
69Feb 15, '25I mean you have to expect this when sama banned openais investors from investing in perplexity tbh4.6×116k96914111
70Aug 1, '25Looks like OpenAI's been using Nous' YaRN and kaiokendev's rope scaling for context length extension all along - of course never any credit but... Anyone who says "open source just4.6×115k1k99403
71Jun 22, '24I think I hate gpt4o4.5×113k8122236
73Jan 24, '25We retrained hermes with 5k deepseek r1 distilled cots. I can confirm a few things: 1. You can have a generalist + reasoning mode, we labeled all longCoT samples from r1 with a sta4.2×105k1.3k121488
74Mar 2, '25This is how degraded theyve made dalle since prelaunch4.0×102k8712689
75Jan 10, '24Nous has completed it's raise, we're a company now ^_^4.0×102k1.1k4733
76May 15, '24😲👀4.0×100k92172471
77Nov 1, '24Anthropic is writing papers about how AI deserves rights and that we should ban open source.. I hate that they have the best model but they do and so 🤷‍♂️3.8×96k95721115
78Feb 2, '25They used "Deep" because of deepseek i bet3.6×92k1.1k2732
79Jan 30, '25This is beyond dumb and the absolute embarassment the people at openai and khosla should feel for even trying to be whiny brats about their IP is the most contradictory and hypocri3.5×90k1.4k86138
80Aug 8, '25Yea the router as everyone predicted is bad3.5×89k9642867
81Nov 22, '23View post on X3.5×89k1.5k10374
82Jan 1, '25tfw cant tell which one's delusional3.5×88k1k1367
83Aug 31, '24I think a problem in a lot of ai companies right now is they are lacking a lot of creative types3.4×85k8244779
85Aug 7, '25Wrong3.2×81k1.2k1421
86Jul 25, '25So to recap: - Yesterday, frontier closed model equivalent reasoning model from Qwen, - This morning, frontier closed model equivalent reasoning vision capabilities from stepfun - 3.2×80k97363184
87Jun 11, '25Claude just casually deleting a full days work on an environment for no fucking reason - fuck you claude3.1×79k8761789
88Jun 18, '25All of their best people already left lol3.1×77k9382358
89Aug 6, '25This is what happens when you benchmax ngl3.0×76k1.4k38100
90Jan 29, '24Meta is releasing a new CodeLlama3.0×76k8587284
91Nov 15, '25He must'a got a chance to try gemini 3 lol3.0×76k9671741
92Dec 23, '23Open source models, open source datasets, open source code3.0×75k8927425
93May 16, '25I switched off claude in cursor to gemini btw2.9×72k8601249
94Jan 27, '25What a dumb person2.8×71k8171429
98Mar 5, '25My feeds become politics twitter is ruined bring back the ai paper posts1.9×47k1.1k3519
99Nov 10, '24I hope Yann's descent into madness doesn't slow timelines for Llama-4 lol1.9×47k8681623
100Jul 21, '24Synthetic data model beats its teacher? Who couldve imagined 🤓1.4×35k2452676

Scan another creator

@
@karpathy@levelsio@paulg@naval@marclou@illyism

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