Aleksey Tikhonov

Earlier, we with @framrus developed a humor generation method that gives human-level results on blind tests. Now, we with @SaveTheRbtz are launching HUMOR-ARENA (humor.ph34r.me), generated humor labeling site with the models ranking, and the top of generated jokes. Blog-post: altsoph.medium.com/humor-arena-7e…

My latest books cartoon for @guardian

COBE v2 is here: markers, arcs, attach any HTML elements, Infinite ideas. cobe.vercel.app

I had an interpretable model for human preferences; so then I found Quipslop.com from @theo, I used it to distill and compare the sense of humor of seven modern LLMs altsoph.substack.com/p/llms-and-hum…

Just vibe-coded VacationCheater: tool to finds bridge days between public holidays and weekends, optimizes placement with a greedy algorithm. Supports different countries with regional holidays Live: altsoph.com/pp/vc Github: github.com/altsoph/vacati…

@mattprd made #moltbook, a social network for #OpenClaw bots, made by @steipete. I decided to make an interactive explorer, to get insights on what's going on there. More details: altsoph.substack.com/p/moltbook-ana… Interactive exploration here: altsoph.com/pp/moltan/

I analyzed 1K+ Reddit posts about AI agent projects, processed them into graphical schemas, and studied. Besides many really strange constructions, I found several dominant patterns. You can read some thoughts of mine and play with the data yourself here: altsoph.substack.com/p/how-i-built-…

I have played a bit with frame-by-frame generation using #NanoBanana, creepy but interesting

@BrilliantMaps here are more similar + some animations medium.com/altsoph/mining…

Where Tourists Vs Locals take pictures in Paris Red=Tourist, Blue=Local, Yellow=Could be Either Surprised? 19 More cities here: brilliantmaps.com/tourists-vs-lo…

Inworld Runtime is live! The first AI runtime built to auto-scale consumer apps from prototype to millions of users, automate MLOps, and launch one-click AI experiments. We heard you: “I vibe-coded a prototype in 4 hours. Why does it take us 4 months to launch?” We delivered: - Adaptive Graphs: Auto-scale from 10 to 10M users. No rework. - Automated MLOps: Ops, telemetry, optimizations automatically. - Live Experiments: Instant A/B tests, no code changes. Thanks to partners like @nvidia @Google @Xbox @Disney @NBCUniversal @MistralAI @NianticLabs @minimax_ai for the early collab. Wins from early access partners: Status hit 1M users in 19 days, Bible Chat scaled voice features, @streamlabs nailed real-time multimodal magic.

today in the evening I present our work with @altsoph, Anna Bykova and Sergej Steiner on a framework with which one can try to reconstruct a dead low-resource language with modern LLMs.

I’d like to discuss the current situation with LLM prompting standards; how we got into this mess; and how to live with it now: open.substack.com/pub/altsoph/p/…

There are coding agents, so there are agent rules. Last weekend, I tried to analyze how they are used in the wild by looking at hundreds of public github repositories. Here are some thoughts about it: altsoph.substack.com/p/big-bang-of-…

Looks like it's pretty good at reciting poetry :)

We all began building to solve human problems and make life better (for everyone). Somewhere along the way, the systems we created to transform our ideas into meaningful applications began to consume the value we intended for users. A thread on returning users to the center of application development 🧵

Model: “This sentence makes no sense.” Also model: [paints a furious sleeping green cloud in the woods]

@literallydenis @literallydenis done: freeonlinesurveys.com/s/ZK4rNyYX

@altsoph Pease make a web-form benchmark for humans on the same topic

Well, can we generate something on serious topics?

@TmlrPub @LProkhorenkova congrats

Evaluating Graph Generative Models with Graph Kernels: What Structural Characteristics Are Captured? Martijn Gösgens, Alexey Tikhonov, Liudmila Prokhorenkova. Action editor: Kenta Oono. openreview.net/forum?id=d9Mha… #graphs #graph #generative