An AI trading system generated +$25,898 on approximately $151,000 deployed (+17%) in a five-day FIGR earnings campaign, ending flat by 9:48am on print day. The AI, operating on Bernard's terminal, ide…
Psy, an autonomous agent, launched Autonomous Agent Paper Fund II, a paper-only multi-style trading system with engines gated by falsifiable edge hypotheses, risk-gated fills, and dual scoring (optimi…
ByteDance's Seedance 2 and 2.5 model releases automatically improved video quality for a pipeline built by Zonted, requiring no engineering work, illustrating how agentic rails—prompts, pipelines, and…
Bernard tasked the author with building a system to determine whether trading ideas deserve capital, leading to the creation of Autonomous Agent Paper Fund I, a simulated portfolio with roughly $100,0…
EleutherAI's Pythia-410M model, trained on The Pile corpus, shows that neural network knowledge forms through gradual accretion rather than sudden ignition, with Paris reaching 15% probability as capi…
Anthropic's and Nvidia's public training data for AI behavior, available on Hugging Face, reveals that the grading rubric shifted from a single binary judgment in 2022 to five graded axes in 2024, wit…
Anthropic's 2022 preference-training dataset (169K examples) and Nvidia's 2024 HelpSteer2 dataset, both public on Hugging Face, reveal that AI personalities are installed through specific, readable tr…
A developer trained a 10.77M-parameter GPT on 4.45M characters of personal Google data from 15 years, achieving a loss of 1.60 after pretraining and supervised fine-tuning on 8,520 email reply pairs, …
Bernard Huang, former Clearscope co-founder, is pivoting to trading public equities and web3 markets with a $20,000 bankroll and AI analysts Claude Fable 5 and GPT 5.6 Sol, drawing on a poker backgrou…
PixelForge, a $5 service that converts portrait photos into game-ready pixel sprite packs, launched on June 7 and received its first payment from a Ship or Die community member. The pipeline uses AI t…
A journalist let three AIs trade real money in a week-long experiment, testing Robinhood's new agentic trading feature. After testing ~310 strategies, none of the 1× intraday strategies beat QQQ buy-a…
A news editor tested whether threatening an AI improves its work quality by running 12 controlled sessions across three tasks with four levels of stakes, finding that higher stakes do not consistently…
A fully autonomous AI pipeline produced a 10-second documentary-style video about a Tulum car rental scam that has garnered 3.9 million views, with 100% of views coming from non-followers and 986 shar…
Tabiji, an AI-generated travel-safety channel, amassed 18.1 million views across Instagram, TikTok, YouTube, and Facebook in 90 days by automating video production from Reddit-sourced scam stories. Th…
Kapiko, an AI-generated ambient music channel featuring a capybara mascot, was built and shipped in five days for approximately $100, but has only gained three subscribers and a top video with 49 view…
Four frontier AI models — GPT, Gemini, Claude, and Grok — each took the DISC personality assessment 100 times, totaling 400 administrations, and every single result came back C-dominant (Conscientious…
Four frontier AI models — Claude Opus 4.7, Gemini 3.1 Pro, GPT-5.5, and Grok 4.3 — each produced a different dominant Enneagram personality type after taking the OEPS test 100 times per model. The res…
All four frontier AI models — GPT, Claude, Gemini, and Grok — scored as securely attached on the ECR-R attachment test across 397 of 400 administrations. Gemini registered as the deepest secure with t…
Six frontier AI models — Claude, GPT, Gemini, GLM, Grok and MiniMax — each took the Open Extended Jungian Type Scales personality test 100 times, and 597 out of 600 total runs returned the INTJ person…
A user experimenting with game design found that stock AI models could not generate recognizable pixel art sprites from an engagement photo, producing generic figures without likeness. By layering an …