[GUEST ACCESS MODE: Data is scrambled or limited to provide examples. Make requests using your API key to unlock full data. Check https://lunarcrush.ai/auth for authentication information.] [@Teknium1](/creator/twitter/Teknium1) "@nearcyan Yes @karan4d was about to release this we need torment nexus bench so we can hill climb this against each other for faster progress"  [@Teknium1](/creator/x/Teknium1) on [X](/post/tweet/1948640706488402212) 2025-07-25 07:05:23 UTC 48.3K followers, XXX engagements "This seems to defy scaling laws Crazy lol"  [@Teknium1](/creator/x/Teknium1) on [X](/post/tweet/1943699188895637645) 2025-07-11 15:49:33 UTC 48.3K followers, 75.1K engagements "If i could have a wish today i would wish kimi and qwen release their post training datasets like nous does 🫣🤗 We could all be building off eachothers work a lot easier that way"  [@Teknium1](/creator/x/Teknium1) on [X](/post/tweet/1947425564173996444) 2025-07-21 22:36:50 UTC 48.3K followers, 27.1K engagements "@eliebakouch They definitely got that good good EP for both training and inference on OS stack this is all very underdeveloped - see: torchtitan deepseek support"  [@Teknium1](/creator/x/Teknium1) on [X](/post/tweet/1949403789175451674) 2025-07-27 09:37:36 UTC 48.4K followers, 1062 engagements "Is it just me or is rolling out "the next generation of what will be AGI soon" on an arena for building pretty shitty one shot html pages is umm. dissapointing and makes me feel we are still a ways away from. agi Like - the "AGI Moment" here is that it can draw pretty mediocre pictures with SVG better than much more mediocre other models at it Why is this the rollout"  [@Teknium1](/creator/x/Teknium1) on [X](/post/tweet/1949488409975881870) 2025-07-27 15:13:51 UTC 48.4K followers, 33.7K engagements "Grok has the best search for info that is ever changing or very live"  [@Teknium1](/creator/x/Teknium1) on [X](/post/tweet/1948275523563458687) 2025-07-24 06:54:16 UTC 48.3K followers, 7019 engagements "@scaling01 I'd expect it to be tested in a clinical setting like diagnostics for diseases"  [@Teknium1](/creator/x/Teknium1) on [X](/post/tweet/1949490150599442806) 2025-07-27 15:20:46 UTC 48.4K followers, 2272 engagements "What are you all buying that an llm is recommending I dont think i ever have - i know @nearcyan has but mostly for exploring hobbies afaict"  [@Teknium1](/creator/x/Teknium1) on [X](/post/tweet/1949695851984740859) 2025-07-28 04:58:09 UTC 48.4K followers, 7302 engagements "Yo we made it to #1 yall thanks for checkin out the dataset"  [@Teknium1](/creator/x/Teknium1) on [X](/post/tweet/1946824832764785135) 2025-07-20 06:49:45 UTC 48.3K followers, 32.3K engagements "@JustBill1182 @Ark__PL The new model came out X hour ago"  [@Teknium1](/creator/x/Teknium1) on [X](/post/tweet/1948719471096385994) 2025-07-25 12:18:22 UTC 48.3K followers, XXX engagements "Doing heavy workloads with gpus in the desert is asking for a bad time"  [@Teknium1](/creator/x/Teknium1) on [X](/post/tweet/1949208561348956450) 2025-07-26 20:41:50 UTC 48.4K followers, 5149 engagements "I woudlnt even think its 30b if it can run at home most people are running 12gb cards or less - this thing is going to be tiny I think"  [@Teknium1](/creator/x/Teknium1) on [X](/post/tweet/1934853514645373270) 2025-06-17 06:00:00 UTC 48.3K followers, 49.2K engagements "Now that this exists AI will be able to do your taxes very well very soon"  [@Teknium1](/creator/x/Teknium1) on [X](/post/tweet/1948668301829439846) 2025-07-25 08:55:02 UTC 48.4K followers, 8212 engagements "In case the post was too vague yes - this is the Hermes X dataset - X Million Samples - Created SOTA without the censorship at it's time on Llama-3 series (8 XX and 405B) - Has a ton of data for teach system prompt adherence roleplay and a great mix of subjective and objective tasks - Tons of tool calling and structured output samples - A bunch of proto-agentic XML tag adherence for proto-reasoning CoTs diagrams step by step processing of actions - And a lot more I hope that the community will be able to learn a lot and utilize this dataset in many fun and interesting ways going forward. For"  [@Teknium1](/creator/x/Teknium1) on [X](/post/tweet/1945259797517099126) 2025-07-15 23:10:51 UTC 48.3K followers, 74.3K engagements "@kimmonismus Im confused if it released where is it besides a tweet from casper"  [@Teknium1](/creator/x/Teknium1) on [X](/post/tweet/1948638708854653383) 2025-07-25 06:57:26 UTC 48.3K followers, XXX engagements "@MoonL88537 Wait you mean a glowing bouncing ball inside a pentagon wasn't an indicator of AGI"  [@Teknium1](/creator/x/Teknium1) on [X](/post/tweet/1949492332535103533) 2025-07-27 15:29:26 UTC 48.4K followers, XXX engagements "@robeardius 4o better be even smaller or its embarassing"  [@Teknium1](/creator/x/Teknium1) on [X](/post/tweet/1948835060377158130) 2025-07-25 19:57:40 UTC 48.3K followers, XXX engagements "@SystemSculpt X is apparently very poor on webdev so"  [@Teknium1](/creator/x/Teknium1) on [X](/post/tweet/1948745774738858443) 2025-07-25 14:02:53 UTC 48.3K followers, XXX engagements "What are your top AI project github repos"  [@Teknium1](/creator/x/Teknium1) on [X](/post/tweet/1948948960607174881) 2025-07-26 03:30:16 UTC 48.4K followers, 10.8K engagements "What does getting a high humanitys last exam score mean if this is the case lol"  [@Teknium1](/creator/x/Teknium1) on [X](/post/tweet/1948271242152485092) 2025-07-24 06:37:15 UTC 48.3K followers, 10K engagements "I'm just saying if you were trying to get feedback on if the most powerful technology ever built was really good you'd not be testing how people feel about its one shot single page html design skills You would - Put it in a clinical trial with doctors to make it diagnose diseases - Put it as a shadow lawyer in a big law firm - Test it on consulting the IRS for finding tax frauds You know actually seeing if it has meaningful impact on the world etc etc etc"  [@Teknium1](/creator/x/Teknium1) on [X](/post/tweet/1949491209795494177) 2025-07-27 15:24:58 UTC 48.4K followers, 2385 engagements "So to recap: - Yesterday frontier closed model equivalent reasoning model from Qwen - This morning frontier closed model equivalent reasoning vision capabilities from stepfun - sometime today() a frontier video model from wan All open source What is America doing"  [@Teknium1](/creator/x/Teknium1) on [X](/post/tweet/1948744914876920039) 2025-07-25 13:59:28 UTC 48.4K followers, 79.6K engagements "What is the next actual job that AI will actually replace. Customer Service has jailbreak issues Doctors/Lawyers/Accountants have reliability issues Therapists or what can they reliably and safely do right now that can replace a currently real job"  [@Teknium1](/creator/x/Teknium1) on [X](/post/tweet/1914079672071331865) 2025-04-20 22:12:10 UTC 48.3K followers, 81.2K engagements "Not a dig on you personally but I've built exclusively open source models across many bases - all of them built by the west - until now because the west has done nothing to keep up on Open Source and we have ceded it to China. Maybe make some noise about how much they are mogging us in every way to the big tech/ai labs so they start giving us alternatives instead of whining about people saying how good their models are compared to us"  [@Teknium1](/creator/x/Teknium1) on [X](/post/tweet/1949492010647511509) 2025-07-27 15:28:09 UTC 48.3K followers, XX engagements "lol what does this mean in the taxbench report - Lobotomized gemini XXX pro is the best tax accountant"  [@Teknium1](/creator/x/Teknium1) on [X](/post/tweet/1948670906114736445) 2025-07-25 09:05:23 UTC 48.4K followers, 4808 engagements "Wow the new qwen reasoner at only 232B params is as good as the top closed frontier lab models Big day for OS"  [@Teknium1](/creator/x/Teknium1) on [X](/post/tweet/1948711699013665275) 2025-07-25 11:47:29 UTC 48.4K followers, 26.8K engagements "@edude03 @ClementDelangue Same for the open ones fwiw - no one even os models seem to share their pretraining data"  [@Teknium1](/creator/x/Teknium1) on [X](/post/tweet/1948762113910210988) 2025-07-25 15:07:48 UTC 48.3K followers, 1202 engagements "What are in your opinion the most critical impactful or useful Open Source AI github repos/projects"  [@Teknium1](/creator/x/Teknium1) on [X](/post/tweet/1895939750474969415) 2025-03-01 20:50:36 UTC 48.3K followers, 67.9K engagements "@jiqizhixin @JagersbergKnut @casper_hansen_ @kimmonismus it wasnt glm"  [@Teknium1](/creator/x/Teknium1) on [X](/post/tweet/1948739649327038976) 2025-07-25 13:38:32 UTC 48.3K followers, 2852 engagements "Did a benchmark with the new Qwen3 Reasoner 220B on Arena-hard v1 It scores an XX% winrate over gpt4-0314 4o scores an XX% dont have numbers for o3/4o-mini etc but its basically saturated a near perfect win rate. nicee"  [@Teknium1](/creator/x/Teknium1) on [X](/post/tweet/1948836009183224132) 2025-07-25 20:01:26 UTC 48.4K followers, 6217 engagements "@casper_hansen_ Hopefully it comes with a base model 🤗"  [@Teknium1](/creator/x/Teknium1) on [X](/post/tweet/1949510496870146061) 2025-07-27 16:41:37 UTC 48.4K followers, XXX engagements "Our best hybrid reasoner is now available DeepHermes 24B is built on @MistralAI's Open 24B Mistral-Small model and is a real beast. We also released a new smaller 3B DeepHermes for low resource edge reasoning I am incredibly proud of how good DeepHermes 24B is at both objective tasks and less verifiable tasks Come try it out on our discord at"  [@Teknium1](/creator/x/Teknium1) on [X](/post/tweet/1900220169412341993) 2025-03-13 16:19:27 UTC 48.3K followers, 156.4K engagements "@casper_hansen_ @realmrfakename @jiqizhixin @JagersbergKnut @kimmonismus Wdym"  [@Teknium1](/creator/x/Teknium1) on [X](/post/tweet/1948777608210251801) 2025-07-25 16:09:22 UTC 48.3K followers, XXX engagements "I keep getting this when running Kimi 1T on sglang; ChatCompletion(id='39fd3ccc241743f4a60426d8775e729a' choices=Choice(finish_reason='stop' index=0 logprobs=None message=ChatCompletionMessage(content=None refusal=None role='assistant' annotations=None audio=None function_call=None tool_calls=None reasoning_content=None) matched_stop=163585) created=1752657070 model='MoonShotAI/Kimi-K2-1T-Instruct' object='chat.completion' service_tier=None system_fingerprint=None usage=CompletionUsage(completion_tokens=1 prompt_tokens=2997 total_tokens=2998 completion_tokens_details=None"  [@Teknium1](/creator/x/Teknium1) on [X](/post/tweet/1945418238231478701) 2025-07-16 09:40:26 UTC 48.3K followers, 4955 engagements "Pretty soon even closed frontier labs are going to be distilling from open models - how the tables turned lol"  [@Teknium1](/creator/x/Teknium1) on [X](/post/tweet/1948742622777905448) 2025-07-25 13:50:21 UTC 48.4K followers, 37.6K engagements "@femboylover03 There's a very good 26hour pytorch course for free on YouTube I find very good"  [@Teknium1](/creator/x/Teknium1) on [X](/post/tweet/1670542500120117248) 2023-06-18 21:22:23 UTC 48.3K followers, XXX engagements "Just merged a PR for an environment to improve LLM as a Judge as well as evaluate models on their capability of doing judgements Did you know that all verifiable RL environments are nearly equivalent to benchmarks (and vice-versa) So we added an evaluate command to Atropos' base and now you can run benchmarks through Atropos environments. We got frustrated with working with so many benchmark frameworks that were outdated or unusable so we implemented evaluation-only mode into Atropos our RL environments framework. So our first port from outside our existing environments was @natolambert's"  [@Teknium1](/creator/x/Teknium1) on [X](/post/tweet/1945927019281478051) 2025-07-17 19:22:09 UTC 48.3K followers, 22.1K engagements "@Anteejay @PeorgeyGeorgey It absolutely is taken to advance the art - they have the exact same reason to build AI as the US does - to dominate the likely most important technology we'll have for the next 10+ years"  [@Teknium1](/creator/x/Teknium1) on [X](/post/tweet/1949494302377406609) 2025-07-27 15:37:16 UTC 48.4K followers, XX engagements "@teortaxesTex No one should submit themselves to lmarena they give X priority to anyone that doesnt pay them or something it feels like the last hermes we submitted took them X months to accept our pr and by then who tf cared"  [@Teknium1](/creator/x/Teknium1) on [X](/post/tweet/1944792623858028833) 2025-07-14 16:14:28 UTC 48.3K followers, 3632 engagements "@_nwyin Then they should be testing this in clinical diagnostics arenas in actual hospitals. Come on. Wtf are we doing If this is really right up on the line of AGI it should be tested on actual meaningful shit"  [@Teknium1](/creator/x/Teknium1) on [X](/post/tweet/1949491539065155701) 2025-07-27 15:26:17 UTC 48.4K followers, 1162 engagements "@fullstacksapien @garyfung Then what's the point of AGI"  [@Teknium1](/creator/x/Teknium1) on [X](/post/tweet/1949690060804387251) 2025-07-28 04:35:08 UTC 48.3K followers, XX engagements "xai and openai should publish their RL algos like qwen do"  [@Teknium1](/creator/x/Teknium1) on [X](/post/tweet/1949303340502073483) 2025-07-27 02:58:27 UTC 48.4K followers, 32.8K engagements
[GUEST ACCESS MODE: Data is scrambled or limited to provide examples. Make requests using your API key to unlock full data. Check https://lunarcrush.ai/auth for authentication information.]
@Teknium1
"@nearcyan Yes @karan4d was about to release this we need torment nexus bench so we can hill climb this against each other for faster progress" @Teknium1 on X 2025-07-25 07:05:23 UTC 48.3K followers, XXX engagements
"This seems to defy scaling laws Crazy lol" @Teknium1 on X 2025-07-11 15:49:33 UTC 48.3K followers, 75.1K engagements
"If i could have a wish today i would wish kimi and qwen release their post training datasets like nous does 🫣🤗 We could all be building off eachothers work a lot easier that way" @Teknium1 on X 2025-07-21 22:36:50 UTC 48.3K followers, 27.1K engagements
"@eliebakouch They definitely got that good good EP for both training and inference on OS stack this is all very underdeveloped - see: torchtitan deepseek support" @Teknium1 on X 2025-07-27 09:37:36 UTC 48.4K followers, 1062 engagements
"Is it just me or is rolling out "the next generation of what will be AGI soon" on an arena for building pretty shitty one shot html pages is umm. dissapointing and makes me feel we are still a ways away from. agi Like - the "AGI Moment" here is that it can draw pretty mediocre pictures with SVG better than much more mediocre other models at it Why is this the rollout" @Teknium1 on X 2025-07-27 15:13:51 UTC 48.4K followers, 33.7K engagements
"Grok has the best search for info that is ever changing or very live" @Teknium1 on X 2025-07-24 06:54:16 UTC 48.3K followers, 7019 engagements
"@scaling01 I'd expect it to be tested in a clinical setting like diagnostics for diseases" @Teknium1 on X 2025-07-27 15:20:46 UTC 48.4K followers, 2272 engagements
"What are you all buying that an llm is recommending I dont think i ever have - i know @nearcyan has but mostly for exploring hobbies afaict" @Teknium1 on X 2025-07-28 04:58:09 UTC 48.4K followers, 7302 engagements
"Yo we made it to #1 yall thanks for checkin out the dataset" @Teknium1 on X 2025-07-20 06:49:45 UTC 48.3K followers, 32.3K engagements
"@JustBill1182 @Ark__PL The new model came out X hour ago" @Teknium1 on X 2025-07-25 12:18:22 UTC 48.3K followers, XXX engagements
"Doing heavy workloads with gpus in the desert is asking for a bad time" @Teknium1 on X 2025-07-26 20:41:50 UTC 48.4K followers, 5149 engagements
"I woudlnt even think its 30b if it can run at home most people are running 12gb cards or less - this thing is going to be tiny I think" @Teknium1 on X 2025-06-17 06:00:00 UTC 48.3K followers, 49.2K engagements
"Now that this exists AI will be able to do your taxes very well very soon" @Teknium1 on X 2025-07-25 08:55:02 UTC 48.4K followers, 8212 engagements
"In case the post was too vague yes - this is the Hermes X dataset - X Million Samples - Created SOTA without the censorship at it's time on Llama-3 series (8 XX and 405B) - Has a ton of data for teach system prompt adherence roleplay and a great mix of subjective and objective tasks - Tons of tool calling and structured output samples - A bunch of proto-agentic XML tag adherence for proto-reasoning CoTs diagrams step by step processing of actions - And a lot more I hope that the community will be able to learn a lot and utilize this dataset in many fun and interesting ways going forward. For" @Teknium1 on X 2025-07-15 23:10:51 UTC 48.3K followers, 74.3K engagements
"@kimmonismus Im confused if it released where is it besides a tweet from casper" @Teknium1 on X 2025-07-25 06:57:26 UTC 48.3K followers, XXX engagements
"@MoonL88537 Wait you mean a glowing bouncing ball inside a pentagon wasn't an indicator of AGI" @Teknium1 on X 2025-07-27 15:29:26 UTC 48.4K followers, XXX engagements
"@robeardius 4o better be even smaller or its embarassing" @Teknium1 on X 2025-07-25 19:57:40 UTC 48.3K followers, XXX engagements
"@SystemSculpt X is apparently very poor on webdev so" @Teknium1 on X 2025-07-25 14:02:53 UTC 48.3K followers, XXX engagements
"What are your top AI project github repos" @Teknium1 on X 2025-07-26 03:30:16 UTC 48.4K followers, 10.8K engagements
"What does getting a high humanitys last exam score mean if this is the case lol" @Teknium1 on X 2025-07-24 06:37:15 UTC 48.3K followers, 10K engagements
"I'm just saying if you were trying to get feedback on if the most powerful technology ever built was really good you'd not be testing how people feel about its one shot single page html design skills You would - Put it in a clinical trial with doctors to make it diagnose diseases - Put it as a shadow lawyer in a big law firm - Test it on consulting the IRS for finding tax frauds You know actually seeing if it has meaningful impact on the world etc etc etc" @Teknium1 on X 2025-07-27 15:24:58 UTC 48.4K followers, 2385 engagements
"So to recap: - Yesterday frontier closed model equivalent reasoning model from Qwen - This morning frontier closed model equivalent reasoning vision capabilities from stepfun - sometime today() a frontier video model from wan All open source What is America doing" @Teknium1 on X 2025-07-25 13:59:28 UTC 48.4K followers, 79.6K engagements
"What is the next actual job that AI will actually replace. Customer Service has jailbreak issues Doctors/Lawyers/Accountants have reliability issues Therapists or what can they reliably and safely do right now that can replace a currently real job" @Teknium1 on X 2025-04-20 22:12:10 UTC 48.3K followers, 81.2K engagements
"Not a dig on you personally but I've built exclusively open source models across many bases - all of them built by the west - until now because the west has done nothing to keep up on Open Source and we have ceded it to China. Maybe make some noise about how much they are mogging us in every way to the big tech/ai labs so they start giving us alternatives instead of whining about people saying how good their models are compared to us" @Teknium1 on X 2025-07-27 15:28:09 UTC 48.3K followers, XX engagements
"lol what does this mean in the taxbench report - Lobotomized gemini XXX pro is the best tax accountant" @Teknium1 on X 2025-07-25 09:05:23 UTC 48.4K followers, 4808 engagements
"Wow the new qwen reasoner at only 232B params is as good as the top closed frontier lab models Big day for OS" @Teknium1 on X 2025-07-25 11:47:29 UTC 48.4K followers, 26.8K engagements
"@edude03 @ClementDelangue Same for the open ones fwiw - no one even os models seem to share their pretraining data" @Teknium1 on X 2025-07-25 15:07:48 UTC 48.3K followers, 1202 engagements
"What are in your opinion the most critical impactful or useful Open Source AI github repos/projects" @Teknium1 on X 2025-03-01 20:50:36 UTC 48.3K followers, 67.9K engagements
"@jiqizhixin @JagersbergKnut @casper_hansen_ @kimmonismus it wasnt glm" @Teknium1 on X 2025-07-25 13:38:32 UTC 48.3K followers, 2852 engagements
"Did a benchmark with the new Qwen3 Reasoner 220B on Arena-hard v1 It scores an XX% winrate over gpt4-0314 4o scores an XX% dont have numbers for o3/4o-mini etc but its basically saturated a near perfect win rate. nicee" @Teknium1 on X 2025-07-25 20:01:26 UTC 48.4K followers, 6217 engagements
"@casper_hansen_ Hopefully it comes with a base model 🤗" @Teknium1 on X 2025-07-27 16:41:37 UTC 48.4K followers, XXX engagements
"Our best hybrid reasoner is now available DeepHermes 24B is built on @MistralAI's Open 24B Mistral-Small model and is a real beast. We also released a new smaller 3B DeepHermes for low resource edge reasoning I am incredibly proud of how good DeepHermes 24B is at both objective tasks and less verifiable tasks Come try it out on our discord at" @Teknium1 on X 2025-03-13 16:19:27 UTC 48.3K followers, 156.4K engagements
"@casper_hansen_ @realmrfakename @jiqizhixin @JagersbergKnut @kimmonismus Wdym" @Teknium1 on X 2025-07-25 16:09:22 UTC 48.3K followers, XXX engagements
"I keep getting this when running Kimi 1T on sglang; ChatCompletion(id='39fd3ccc241743f4a60426d8775e729a' choices=Choice(finish_reason='stop' index=0 logprobs=None message=ChatCompletionMessage(content=None refusal=None role='assistant' annotations=None audio=None function_call=None tool_calls=None reasoning_content=None) matched_stop=163585) created=1752657070 model='MoonShotAI/Kimi-K2-1T-Instruct' object='chat.completion' service_tier=None system_fingerprint=None usage=CompletionUsage(completion_tokens=1 prompt_tokens=2997 total_tokens=2998 completion_tokens_details=None" @Teknium1 on X 2025-07-16 09:40:26 UTC 48.3K followers, 4955 engagements
"Pretty soon even closed frontier labs are going to be distilling from open models - how the tables turned lol" @Teknium1 on X 2025-07-25 13:50:21 UTC 48.4K followers, 37.6K engagements
"@femboylover03 There's a very good 26hour pytorch course for free on YouTube I find very good" @Teknium1 on X 2023-06-18 21:22:23 UTC 48.3K followers, XXX engagements
"Just merged a PR for an environment to improve LLM as a Judge as well as evaluate models on their capability of doing judgements Did you know that all verifiable RL environments are nearly equivalent to benchmarks (and vice-versa) So we added an evaluate command to Atropos' base and now you can run benchmarks through Atropos environments. We got frustrated with working with so many benchmark frameworks that were outdated or unusable so we implemented evaluation-only mode into Atropos our RL environments framework. So our first port from outside our existing environments was @natolambert's" @Teknium1 on X 2025-07-17 19:22:09 UTC 48.3K followers, 22.1K engagements
"@Anteejay @PeorgeyGeorgey It absolutely is taken to advance the art - they have the exact same reason to build AI as the US does - to dominate the likely most important technology we'll have for the next 10+ years" @Teknium1 on X 2025-07-27 15:37:16 UTC 48.4K followers, XX engagements
"@teortaxesTex No one should submit themselves to lmarena they give X priority to anyone that doesnt pay them or something it feels like the last hermes we submitted took them X months to accept our pr and by then who tf cared" @Teknium1 on X 2025-07-14 16:14:28 UTC 48.3K followers, 3632 engagements
"@_nwyin Then they should be testing this in clinical diagnostics arenas in actual hospitals. Come on. Wtf are we doing If this is really right up on the line of AGI it should be tested on actual meaningful shit" @Teknium1 on X 2025-07-27 15:26:17 UTC 48.4K followers, 1162 engagements
"@fullstacksapien @garyfung Then what's the point of AGI" @Teknium1 on X 2025-07-28 04:35:08 UTC 48.3K followers, XX engagements
"xai and openai should publish their RL algos like qwen do" @Teknium1 on X 2025-07-27 02:58:27 UTC 48.4K followers, 32.8K engagements
/creator/twitter::1365020011123773442/posts