This week, every week

My agents continue to work on my data layer, which is purpose built for agentic finance usage. I've discovered a new area frontier models suck at - estimating hardware resource usage - since I've been pushing my system to the limits with my blockchain work. (Turns out running a node, multiple data jobs, as well as my other health use cases causes a huge headache of hardware resource contention across NVME, HDD, CPU and GPU!) Good thing is I'm working w/ a friend who has site reliability experience to do a v0 for wut I think is potentially an entirely new infra layer. I'm thinking of creating an eval, perhaps on a dataset announced that covers one year of LLM inference metadata traces with 6.12 billion requests, aimed at real-world serving workload understanding, system design, and infrastructure optimization. It's been interesting to see the rise of agentic trading terminals - for the crypto majors and public equities, but not for memecoins as the onchain landscape is still a dark forest (see Pump.Fun rewards to scammers as an example!) I'm experimenting with new ways of real time scam detection as well ;)

I'm happy to be back in grind mode - locked myself in my house, not talking to anyone except family + a few close friends. I'm building my own (enriched for trading) Solana + Robinhood memecoin trading data tables, with some nice surprises and additions ;) I'm inspired by the finetuned superforecasting models and considering exploring that for the prediction market work. Shipped a Jev bookmark yesterday. Also want to spin off some stuff I've done before (my health dashboard work) as a Muse connector since I've been really into my fitness lately (body recomp successful!) I bookmarked a reminder that the important stuff is often what we didn't have time for, and am spending a lot of time thinking about how and where I'm allocating my time. Also buying more memory and storage for my GPU, it really is THE bottleneck. I baked a flourless chocolate cake w/ chocolate from Taiwan + Ecuador/Portugal to celebrate HYPE ATH!

I attending a wonderful wedding in Berkeley and am enjoying the hottest week in SF. I'm running a Robinhood archival node & working on onchain entity resolution for Robinhood and Solana. It's been fascinating to follow Robinhood on chain activity, though it's become increasing adversarial IMO due to lack of gud data as well as more complex onchain iterations. I've also been manually trading political prediction markets (Ossoff FTW!) and am getting ready to begin more programmatic execution. I do think there will be rotations from Solana memes to Robinhood (& vice versa) because of PVP activity and lower liquidity. I've been pleased to meet people in SF who are interested in AI model psychology and impact on human psychology. Have been debating whether AI voice models will ever be gud - they need more params than video & audio, and are the clearest expression of human thinking.

I ran a checker-backed Alpha Challenge evaluation of frontier models solving public blockchain puzzles, which showed that while the current class of frontier and OSS models can write protocol code, they can't reliably navigate onchain or analyze complex onchain activity - there's a capability cliff that I believe can only be solved through better training data or domain-specific fine tuning. That failure increasingly makes me think the path to ASI is a data and environment problem disguised as a model problem. I've been fascinated to see that we're now entering a stage where analyzing model psychology and sociology is critical, since agents now have baseline behavior and functional decision-making processes that are heavily inspired by but systematically different from humans.

Expanded data collection to Kalshi (as a holdout) and now have 60 loops - including 7 autoresearch loops mining diff strats. I rebuilt my personal site around a weekly note and reading shelf, then wired subscriptions, the archive, and an automatic welcome digest. The voice work has a weekly evaluation for measuring how close the model gets to a publishable note without intervention (this week only 3x)! This connects with what I was reading about bottlenecks: the last constraint is often emotional rather than technical. Another reading called out how avoiding decisions can become an avoidance of responsibility. It made me want to choose the next concrete step instead of preserving every possibility.

I'm building an insider trading detector for Polymarket using wallet forensics by running 2 auto research loops - 1) to find novel "statistically improbable" strategies and 2) to backtest the best copy-trading strategy for these wallets! It's been very fun to coordinate multiple agents across my entire compute network (GPU + Mac, frontier + local models). Version 12 of my writing model, a Qwythos-9B fine-tune, is now live, drafting in my voice while I continue to judge its output blind. I've been reading a lot about reinforcement learning (which IMO will be commoditized!) and believe that fine tuning / pre-training small, fully owned open source models tailored to proprietary workflows is the future of enterprise AI.