Simon Solotko

Our new, @tiriasresearch public forecast tool looks at ChatGPT and LLMs pursuing AGI through 2030. Want to understand the potential for future versions of ChatGPT, Gemini, and Llama? Read more and access the public tool on Linkedin: lnkd.in/geA5bdzx

I just backed Arcade Earth: Rise of Vector on @Kickstarter kickstarter.com/projects/dr-ca…

@TiriasResearch forecasts a 115x growth in generative AI token generation by 2030, reaching a staggering 77 quadrillion tokens. Read more about it in @Forbes.com forbes.com/sites/tiriasre…

So I thought this sounded cheesey at first but after watching the video this looks like a pretty cool project 🤔 👉kickstarter.com/projects/rpgme…

⭐️THE RPGme KICKSTARTER IS LIVE! ⭐️ Thank you all so much for patience and support! Go here to check out our Kickstarter! 👉 kck.st/4iKxfCN 👈 RETWEET and TAG a friend below and we'll pick 10 random duos to be put in their own video game! 🚀🚀🚀

I just backed RPGme | Create & Gift A Personalized Retro RPG Adventure on @Kickstarter kickstarter.com/projects/rpgme…

Explore the roadmap of models like ChatGPT and the journey toward AGI with the now Public Generative AI LLM Forecast Tool from Tirias Research. Visualize how large language models are evolving in size, complexity, and computational requirements. linkedin.com/posts/solotko_…

@simonw @simonw our model looks at data center power per token through time and the top line (it's a blend of video, images and tokens) and is public at tiriasresearch.com/research I have the breakouts for tokens and images if anyone is interested.

Has anyone conducted studies that compare the energy usage of many people running local, personal LLMs to that of many people sharing access to much more power-hungry hosted LLMs?

@DrJimFan If you are interested in conjecturing on its size and the compute requirement for the reasoning step, check out our LLM Forecast Tool. linkedin.com/feed/update/ur…

OpenAI Strawberry (o1) is out! We are finally seeing the paradigm of inference-time scaling popularized and deployed in production. As Sutton said in the Bitter Lesson, there're only 2 techniques that scale indefinitely with compute: learning & search. It's time to shift focus to the latter. 1. You don't need a huge model to perform reasoning. Lots of parameters are dedicated to memorizing facts, in order to perform well in benchmarks like trivia QA. It is possible to factor out reasoning from knowledge, i.e. a small "reasoning core" that knows how to call tools like browser and code verifier. Pre-training compute may be decreased. 2. A huge amount of compute is shifted to serving inference instead of pre/post-training. LLMs are text-based simulators. By rolling out many possible strategies and scenarios in the simulator, the model will eventually converge to good solutions. The process is a well-studied problem like AlphaGo's monte carlo tree search (MCTS). 3. OpenAI must have figured out the inference scaling law a long time ago, which academia is just recently discovering. Two papers came out on Arxiv a week apart last month: - Large Language Monkeys: Scaling Inference Compute with Repeated Sampling. Brown et al. finds that DeepSeek-Coder increases from 15.9% with one sample to 56% with 250 samples on SWE-Bench, beating Sonnet-3.5. - Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters. Snell et al. finds that PaLM 2-S beats a 14x larger model on MATH with test-time search. 4. Productionizing o1 is much harder than nailing the academic benchmarks. For reasoning problems in the wild, how to decide when to stop searching? What's the reward function? Success criterion? When to call tools like code interpreter in the loop? How to factor in the compute cost of those CPU processes? Their research post didn't share much. 5. Strawberry easily becomes a data flywheel. If the answer is correct, the entire search trace becomes a mini dataset of training examples, which contain both positive and negative rewards. This in turn improves the reasoning core for future versions of GPT, similar to how AlphaGo’s value network — used to evaluate quality of each board position — improves as MCTS generates more and more refined training data.

Avid Union is a @Kickstarter OG so great to see them back @AVIDUNION makes you more awesome - check out The Shield 2.0 Urban Tactical Soft Shell Jacket kickstarter.com/projects/entho…

@Anticleric @Thrilluwu It's nonstupid - thx for sharing.

This new video from @Thrilluwu is relevant to this thread, it speaks more about overall OS UI design rather than games, but there is a lot of crossover and the concepts are sound. youtu.be/Fhlw88_Beu4?si…

#VR Design Manifesto AKA holodeck program guidelines [a work in progress] - no fail states - minimize user frustration - maximize user empowerment - maximize user awe - allow user to set their own goals - don't put words in user's mouth - don't push the user around

Just announced the availability of our AGI Forecast Model which provides LLM model size and performance projections for OpenAI’s ChatGPT, Google Gemini, and platform capabilities available to open source models like Facebook Llama through 2028 - Contact us to learn more!

Our recently announced AGI Forecast Model places OpenAI’s ChatGPT4o as their first model using NVIDIA H100 for both training and inference benefiting from improved performance/lower TCO-enabling OpenAI to offer API access at a lower price. Contact us to learn more!

The impact of #GenAI will be huge, but none more so than on the tech industry itself. @tiriasresearch examines the impact on GenAI in its FTCO service and in the most recent article on @Forbes.com forbes.com/sites/tiriasre…

Jensen: Blackwell is the name of a platform. Blackwell compared with Hopper. Two die are abutted to form Blackwell chip, making it larger than a reticle size. #GTC2024

Summary slide of Blackwell. #GTC2024

Where AI is going: understanding is multimode. #GTC2024 If there's patterns, we can understand it.

RT+Follow+Like to enter a giveaway for a Glyph prototype unit! Will randomly pick a winner Oct 30th and we will send it right away! Kickstarter campaign ends Nov 1st: limitlabs.com/GlyphKickstart… Also we will be at Big House next weekend at @RectangleCorner 's booth, come try it!

We sent a #GlyphController over to @KevinKenson for some quality hands-on time - check out the full video below! youtu.be/6eq4KLmcPDs

Introducing Glyph: a leverless fightstick with swappable layouts to support platform fighters, traditional fighters, retro games, and more. Kickstarter is now live, starting at $259.99 limitlabs.com/GlyphKickstart…