• Home
  • Technology
  • Gaming
  • Entertainment
  • World & Business
  • Science
  • Sports
  • AI
HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
    • HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
    • Home
    • Technology
    • Gaming
    • Entertainment
    • World & Business
    • Science
    • Sports
    • AI
    AI

    Yash Patil Calls Specialized Models the Future of AI

    Applied Compute CEO and other founders post about narrow AI models for business tasks.

    PG
    OK
    TL
    29 Sources, 29d ago, first seen 29d ago

    TLDR

    Yash Patil, Applied Compute CEO and former OpenAI engineer, posted that specialized models are the future. Shopify CEO Tobi Lütke replied that finetuned tiny models work well with self-improving flywheels, noting one case where such a model outperformed a larger general system on a buyer profile task. Moondream CTO Vikhyat K. offered an inference engine for Qwen3.5 0.8B on H100s. DatologyAI CEO Ari Morcos added that business use cases are narrow and pre-specified, giving focused models an edge in cost and control. LangChain and others shared related posts on open weights models and agent traces.

    Combined views

    2.2M

    29 Sources, first seen 29d ago

    Combined views

    2.2M

    29 Sources, first seen 29d ago

    14.7K likes
    14.7K likes
    495 comments
    11.8K saves
    1.7K reposts
    495 comments
    11.8K saves
    1.7K reposts

    Sentiment

    Positive91.5%8.5%Negative

    Summary

    Sentiment

    Positive91.5%8.5%Negative

    Many accounts welcomed tuning smaller specialized models for specific tasks because they can match frontier performance at far lower cost, while a few noted potential robustness tradeoffs.

    Based on 139 sentiment-bearing replies from 129 accounts across 10 conversations.

    Summary

    Many accounts welcomed tuning smaller specialized models for specific tasks because they can match frontier performance at far lower cost, while a few noted potential robustness tradeoffs.

    Based on 139 sentiment-bearing replies from 129 accounts across 10 conversations.

    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet

    29 Sources

    @BryanOffutt@levie The lab models are an encyclopedia, which is perfect for the consumer use case. Little bit of info on any topic. Most business tasks require a textbook. Smaller, more focused volume.
    @tobiTraining tiny models for special purpose use cases works so incredibly well if you have a great self improving recursive flywheel. Shopify ML team is on fire. finetuned 0.8b model beats GPT 5.6-sol xhigh in this very specialized task.
    @omarsar0Early signs of a new era of customized frontier intelligence. It will be epic when the rest of the industry catches up on the potential and exponential applications of custom models. Owning your intelligence stack is not just about ownership; it's how you stay relevant as a future company operating at the frontier of intelligence.
    @arimorcosRT @levie: Now that the base open weights AI models are getting far better, and post training infra is becoming more mature and commerciali…
    @FireworksAI_HQOur partners @LangChain generate billions of tokens of agent traces a day, their richest signal on how real users react. Judging every one with a frontier closed model like GPT-5.5 or Opus is too costly, so they fine-tuned a Qwen base model on Fireworks that matches it at up to 100x lower cost. Build your own frontier: https://fireworks.ai/training
    @sundeepYup.
    @MParakhinAnd the prompt is fully gisted as well! We do do cool things here :-)
    @LangChainRT @FireworksAI_HQ: Our partners @LangChain generate billions of tokens of agent traces a day, their richest signal on how real users react…
    @lateinteractionreminds me of when Shopify folks (@kshetrajna) saved $5M using DSPy - the company seems to be really good at this kind of thing! https://www.youtube.com/watch?v=bxToahwOVpY
    @vikhyatk@tobi looks like you're running Qwen3.5 0.8B on H100s here? we recently shipped an inference engine that runs it really fast, in case that's of interest to you! https://moondream.ai/blog/photon-2-launch

    29 Sources

    @BryanOffutt@levie The lab models are an encyclopedia, which is perfect for the consumer use case. Little bit of info on any topic. Most business tasks require a textbook. Smaller, more focused volume.
    @tobiTraining tiny models for special purpose use cases works so incredibly well if you have a great self improving recursive flywheel. Shopify ML team is on fire. finetuned 0.8b model beats GPT 5.6-sol xhigh in this very specialized task.
    @omarsar0Early signs of a new era of customized frontier intelligence. It will be epic when the rest of the industry catches up on the potential and exponential applications of custom models. Owning your intelligence stack is not just about ownership; it's how you stay relevant as a future company operating at the frontier of intelligence.
    @arimorcosRT @levie: Now that the base open weights AI models are getting far better, and post training infra is becoming more mature and commerciali…
    @FireworksAI_HQOur partners @LangChain generate billions of tokens of agent traces a day, their richest signal on how real users react. Judging every one with a frontier closed model like GPT-5.5 or Opus is too costly, so they fine-tuned a Qwen base model on Fireworks that matches it at up to 100x lower cost. Build your own frontier: https://fireworks.ai/training
    @sundeepYup.
    @MParakhinAnd the prompt is fully gisted as well! We do do cool things here :-)
    @LangChainRT @FireworksAI_HQ: Our partners @LangChain generate billions of tokens of agent traces a day, their richest signal on how real users react…
    @lateinteractionreminds me of when Shopify folks (@kshetrajna) saved $5M using DSPy - the company seems to be really good at this kind of thing! https://www.youtube.com/watch?v=bxToahwOVpY
    @vikhyatk@tobi looks like you're running Qwen3.5 0.8B on H100s here? we recently shipped an inference engine that runs it really fast, in case that's of interest to you! https://moondream.ai/blog/photon-2-launch