25 YC startups that have trained their own AI models

Nural Research -

Welcome to Nural's newsletter focusing on how AI is being used to tackle global grand challenges.

Packed inside we have

  • 25 YC startups that have trained their own AI models
  • Data acquisition strategies for AI-first start-ups
  • and [Anthropic] Many-shot jailbreaking: Eliciting harmful responses from modern large context LLMs

If you would like to support our continued work from £2/month then click here!

Marcel Hedman

Key Recent Developments

Reaching LLaMA2 Performance with 0.1M Dollars

https://research.myshell.ai/jetmoe

What: Researchers have achieved on par performance with Meta's released LLaMA models while only spending $100k. The ability to achieve state of the art (SOTA) capabilities with small finite budgets opens the possibility for researchers and smaller startups to contribute towards AI advancement alongside the mega AI labs.

2024 Machine Learning AI and Data Landscape

Full Steam Ahead: The 2024 MAD (Machine Learning, AI & Data) Landscape
This is our tenth annual landscape and “state of the union” of the data, analytics, machine learning and AI ecosystem. In 10+ years covering the space, things have never been as exciting and promising as they are today. All trends and subtrends we described over the years are coalescing: data ha
Matt TurckJeff Evernham

What: This article contains a comprehensive overview of the global AI and data landscape. You will note ML & AI and its complete vertical stack is just one segment in a broader interconnected web spanning analytics, infrastructure and consulting players.

25 YC companies that have trained their own AI models

What: Last week YCombinator, the world's premier startup incubator, had its demo day. During this day, startups showcase their efforts from their period on the programme and understandably, many of these companies had an AI focus.

The attached list showcases a number who have not only incorporated existing AI techniques, but have driven towards full ownership of their own trained models. Perhaps a good step towards defensibility or a wasted effort in a world where new models and architectures are constantly released?

AI Ethics & 4 good

🚀 [Anthropic] Many-shot jailbreaking: Eliciting harmful responses from modern LLMs

🚀 Generation of synthetic whole-slide image tiles of tumours from RNA-sequencing data via cascaded diffusion models

Other interesting reads

🚀 Governing AI agents: Limitations of conventional solutions to agency problems in an AI world

🚀 Data acquisition strategies for AI-first start-ups

Papers

🚀 Visualization-of-Thought Elicits Spatial Reasoning in Large Language Models

🚀[Apple] ReALM: Reference Resolution As Language Modeling (incorporating on-screen context)

🚀[Dataset] Text --> SQL

🚀 Foresight—a generative pretrained transformer for modelling of patient timelines using electronic health records: a retrospective modelling study

Cool companies found this week

Health

Deepgram - API based solution for speech-to-text transcription and text-to-speech.

Weather forecasting

Atmo - AI driven weather forecasting, replacing traditional physics engines.

Best,

Marcel Hedman
Nural Research Founder
www.nural.cc

If this has been interesting, share it with a friend who will find it equally valuable. If you are not already a subscriber, then subscribe here.

If you are enjoying this content and would like to support the work financially then you can amend your plan here from £2/month!