Albatross

GLM 5.2 vs DeepSeek V4: share of tokens on OpenRouter GLM 5.2 has had a very fast uptake relative to most open-weight launches

Gemma 4 is a very good small model that punches above it's weight class Gemma is a 31B model that is as good as other very large MoE models It's the best in the world for it's size 👏👏

This didn't receive the attention it deserved. They pre-trained this model completely peer 2 peer, no data-centers. Everything was done over a permissionless network, I have tried the model, it's honestly not a good LLM but that's beyond the point. We NEED this, we NEED an alternative. - Download OpenCode - Download Pi - Pay for OpenSource - Share your AI sessions - Learn to do RL We can't be at the mercy of ANY lab. arxiv.org/abs/2603.08163

Qwen3.5, MiniMax-M2.7 are incredible acts of kindness that I don't think will be with us from so much longer. Here's my update for you. > I have 20 GPUs at full utilisation right now. All these getting cooooompressed, no synthetic data All runs will be done in 9 days, if I don't get a catastrophic failure - REAP for: - GLM-5 - Qwen3-next-coder - Qwen3.5-122B - Qwen3.5-plus-397b - Browser-use - CUDA - Terminal-use - Coding - Math - Agentic trajectories - 30% my personal chat session history I am also removing refusals inspired by Prism. So no more I can't do this I can't do that blah blah Inference for local AI - Qwen3.5-262B-REAP - I've been using it exclusively in Parchi, perfect 100 tokens/s & 0 errors very good at browser use ----------------- Secret - Qwen3.5-27b - you will see when i'm done Targeting the following hardware levels: With full context 200-256k context in vllm, sglang, llama.cpp, exllamav3, and if people help MLX 16-32 GB - Qwen3.5-27b 32-48 GB - Qwen3-coder-next 48-128 GB - Qwen3.5-122B 128-256 GB - Qwen3.5-Plus-397B 196-512 GB - GLM-5.* I am training them on 22,000 samples at 16k context 352M of custom selected calibration datasets. My hope is to make the highest quality multimodal LLM compressions for this year. 20 GPUs running in parallel for the next 10 days - 8x H100s - Qwen - 4x B200s - GLM-5.* - 8x 3090s - Testing Once MiniMax-M2.7 is online 4 more GPUs will get to work.

Python is mind-boggling for finance. Case in point: There's a Finance database of 300,000 tickers. Available 100% for free:

Automate your trading strategies in Python How to build your first trading bot: (a thread)

it was a good idea to give computers to humans. it’s a better idea to give computers to computers so that they can create the same outputs we do on a computer for our work. that’s the core idea of perplexity computer in a nutshell.

Preview of personal portfolio integration and analysis coming soon to Perplexity Finance. The team is testing it this weekend.

AI agents are trapped in chat bubbles. @usecarson breaks free by generating custom, task specific interfaces with enterprise-grade security. Sign up today at usecarson.com. Congrats on the launch, @sidharthkmenon, @_ketan0, @alex_iansiti, and @milanb_123! ycombinator.com/launches/PXG…

The way startups are built has shifted quickly. We're excited about a range of startup ideas for AI-native companies that can now be built faster, cheaper, and with more ambition than ever. ycombinator.com/rfs

.@usecardboard is an agentic video editor where you upload raw footage and edit it with AI. Unicorns, media agencies, and creators with millions of followers are already using Cardboard. It lives in your browser and helps you go from a sea of raw clips into a finished video in minutes. Congrats on the launch @sxmawl and @ishandeveloper! ycombinator.com/launches/PM3…

YC Demo Day median dilution for small seed is 11% (top 25% is 7%), not 18% for non-YC YC Demo Day median dilution for big seed is 13%, not 21% to 23% for non-YC You do it all faster: YC funds you get pre-seed and in 4 months you have your seed, and you're off to the races, no BS, plus you get the network, the media, the community, and 1:1 partner support for the life of the co. YC pays for itself and then some. There's nothing like it for founders. (There's nothing like it for investors too, median MOIC is 5x for perennial investors at Demo Day, bottom quartile is 3.3x)

Slash, @slashapp, just crossed $150M in annualized revenue profitably. We went from $2M -> $150M in 24 months making us the fastest-growing business banking* platform of all time. 700 word post on 4 guiding principles that got us here. This cost us >$10M dollars to learn... (Bookmark this) I'll cover: • Picking the right market (where 99% of founders go wrong) • Why revenue is the ONLY business metric that matters • Why market saturation is fake • What every founder does day to day that they shouldn't 1. Attack “small” markets: Startup founders - myself included - gravitate towards working on companies that have huge upside. Here's the problem: it’s difficult to find aggressive product-market fit / build a differentiated product if you don’t sell ONE offer to ONE person. I'll repeat: one offer, to one person Examples: • PayPal didn't start by trying to own 70% of online payments they started with payment processing just for eBay merchants. • Uber started as black cars for rich SF people. Ask yourself, what am I selling and to who? If you're selling more than one thing to more than one person, in the beginning, you're not niche enough. Slash started by building a better credit card for SNEAKER RESELLERS. Ridiculously niche. And that tiny wedge alone got us to $5M ARR in 11 months. Once you dominate the niche, you earn the right (and the cash) to expand outward. We STILL go after “small” verticals because our competitors are too arrogant to do it. We walk in and own them. 2. Revenue is the only metric that matters. Everything else is cope. If your revenue isn't growing, nothing else matters. Revenue gives you two things: A) Money to redeploy. (Obvious.) B) Momentum. A team that’s winning wants to work harder. A team that’s losing checks out. To become a unicorn, you have to outwork everyone else. To outwork everyone else, you need morale. To get morale, you need wins. To get wins, you need revenue. Everything ladders back to one thing: Sell more, sooner. Drive sales and demand → everything else falls into place. 3. “Market saturation” is fake. When starting Slash, everyone told us we'd never be successful because Ramp, Brex, and Mercury were already worth > $10bn. The reality is that fintech is only 5% penetrated. 95% of business deposits and corporate card spend still runs through the legacy banks. Many markets are similar to B2B fintech. They can “feel” settled because there is a sexy startup that everyone’s heard about, but dinosaurs have all the rev share. There's always a way to find your wedge. 4. 99% of founders do the wrong thing at the wrong time When you start your role as the founder is to do EVERYTHING. And you should outsource nothing. Example: If you run an ecom business you should write film and edit EVERY single script. If you're a CTO you should write every line of code. Biggest 🚩in an early stage founder is someone who says they need to "outsource to an expert". No. You ARE the expert or you become one. Founders who outsource early are lazy. When you grow this needs to change rapidly. >10M ARR you need to SHIFT fast. Your role as the founder should be to bring in people competent enough to deliver on all of your initiatives. There is simply too much to do and it won’t be possible for you to brute force your way out of every problem. We're winning because 65% of our team is on the spectrum. We have savant engineers who this year alone have, shipped treasury, Stablecoin Payments, check deposits, SWIFT, Global USD, accounting automations, a completely new interface, and more. We have a world-class GTM and ops team. Because of it, we blow our competitors out of the water when it comes to revenue / employee, payment volume / employee, and other efficiency metrics. -------- If you have read this far, thank you. I got told countless times Slash would never be anything. We want Slash to be the first trillion dollar fintech company in the world. At our current growth rate we'll hit 1 billion dollars in revenue run rate in 18 months and 100 billion in 7 years. We’re giving it our all to accelerate our growth rate and hit these metrics even faster.