Falcon is now the first permissive open-source model to "beat" LLaMA, which is still for research purposes. There's been a lot of buzz around open source models, but we haven't really seen them in prod at the scale of a GPT-4 (which most people I talk to use out of convenience.)
@mattlynley
-
Falcon 40B License Change: Open Source LLM Update
By
–
Today's issue of Supervised is about the bizarre week for Falcon 40B, the latest open source LLM that was just placed under a more permissive open source license. It launched last week under a license that required royalties past a certain level of revenue—which changed Wednesday
-
Community Stewardship and Strategic Partnerships in AI Ecosystems
By
–
The downside: you aren't really in control of your own community, and you could easily chase them off. But you can also steward and try to nudge in the right direction, which is what Dbt, HF, etc excel at. And inevitably cos that might build a Dbt or HF end up just partnering.
-
Community-Driven Growth: How AI Startups Compete With Big Tech
By
–
Some of the most successful recent companies—Dbt Labs, Hugging Face, Midjourney—have been able to grow and scale because they have built such a strong community. That's given them a window to scale up with less risk of getting smushed by an incumbent (like a Google).
-
AI Startups Build Competitive Moats Through Fan Loyalty
By
–
I'm sure everyone remembers the (groan-inducing) "data is the new oil" phrase thrown around in the 2010s. Well, modern AI and big data startups *don't* have that and have to find something else to build a mini Moat to buy time for them to scale: Fans.
-
Moats in AI: Why Technical Prowess and Data Aren’t Enough
By
–
Today's (post-holiday for which there are no holidays when starting something new) issue of Supervised is about capital-M Moats and how they fit into big data and modern AI. Specifically, technical prowess and first-mover are useless, and most startups don't have capital-d Data.
-
Streamlit Acquihire: Sridhar Joins as Leadership Expands
By
–
One small wrinkle: a big driving factor for the acquihire was getting Sridhar in the door. Adrien Treuille, the founder of Streamlit, has lately assumed more oversight for broader ML beyond just Streamlit. So we will have to see how the or chart plays out.
-
Snowflake’s Data Advantage for AI Model Fine-Tuning
By
–
For Snowflake, it gets a little interesting. They already host so much company data—the kind of data that’ll get used for fine-tuning—and just need a team that has experience with model dev/production to figure out that workflow.
-
Consumer AI Adoption Challenges vs Enterprise Suitability
By
–
As fast as AI is developing, consumers are still very much locked in their ways and it’s an uphill battle to change core behaviors. And the tools they (and many future consumer AI startups) built were uniquely suited to enterprise use cases.
-
Neeva’s AI Search Failed Due to Switching Costs
By
–
Neeva went for an AI-powered experience earlier this year, but that wasn’t the issue—people were already duct taped to google and bing, and the cognitive switching cost created an enormous barrier to growth, even if Neeva was able to get to AI powered results faster.