🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

[guy who owns a ton of Pokémon cards]: the United States needs a federal Pokémon card reserve

The biggest question today is whether AI lowering the barrier to content creation helps or hurts incumbents like Roblox On the one hand, Roblox owns game distribution for kids under 16 On the other hand, democratized creation opens the opportunity for a new creator tool to

Very excited to announce our investment in Ceramic! In a world of rapid model distillation (best highlighted by Deepseek), unsupervised reinforcement learning, and potentially broken pre-training scaling laws, we believe there will be a new wave of model democratization and

This is the bull case for model agnostic apps (Perplexity/Cursor/etc) - by routing to the appropriate model given the task they both have both a better cost structure + actually outperform since different models are good at different things

Clearly AI is dramatically impacting search behavior, yet it hasn’t shown up in Google’s earnings. Hypothetically, it could be that: -These types of queries (likely informational) weren’t very monetizable before AI anyways AND/OR -Google is increasing it’s ad load

Big TAM for voice agents in dealerships

This is why many high PMF AI widgets and AI services rollups will get crushed by competition in the long-run. The exception are: 1) High trust/risk industries: when customers find something they know works, they stick with it 2) High switching costs: the more products you

Can’t bribe an AI customer support agent

i do think the future of work is like starcraft or age of empires. you have 200 microagents you’re directing to fix problems, gather information, reach out to people, design new systems, etc.

Don't use LoRA if you care about scaling test-time compute (i.e. reasoning use cases)! We tested LoRA RL vs full RL - full post-training was 40% cheaper, 12X faster, AND 50% better than LoRA on out-of-distribution data. Details + link to full blog below! 1/

I wonder how much of LLM’s inability to run even the most simple business is because most business advice on the internet is bad

It’s easy to fine-tune small models w/ RL to outperform foundation models on vertical tasks. We’re open sourcing Osmosis-Apply-1.7B: a small model that merges code (similar to Cursor’s instant apply) better than foundation models. Links to download and try out the model below!

GlobalFoundries agrees to acquire MIPS, a developer of RISC-V-based solutions and IP, with the deal expected to close in H2 2025 (Anton Shilov / Tom's Hardware) tomshardware.com/pc-components/… techmeme.com/250708/p27#a25… x.com/Techmeme/statu…

Probably the best AI video ad I’ve seen Kalshi can’t outspend Draftkings/Fanduel/others so props to Tarek Mansour for finding GTM Alpha by taking risks on AI-generated video when no one else will And props to PJ Ace and team for telling a great story that takes