Physics Researcher Uses Claude to Trade Stocks for a Week, Sparks Debate on AI Value

·1 views

A physics researcher with no finance background connected Anthropic's Claude AI to a live US brokerage account and ran an automated trading experiment for one week using $500. Following a conservative, rules-only strategy — long equities, no margin or shorting — the Claude-managed account returned 2.20%, marginally outperforming benchmark ETFs VOO and QQQ over the same period. The researcher ran a parallel manually managed account, which exhibited classic retail investor pitfalls such as overtrading and emotional decision-making, while the AI account strictly followed the plan without deviation. The experiment, which drew over 374 comments in Chinese AI developer communities, has shifted debate away from raw returns toward a deeper question: whether an LLM like Claude adds anything meaningful over a conventional rules-based Python script, given that free algorithmic brokerage APIs have been widely available for years. The author cautions that one week of data is not a conclusive result, and the full strategy logic has not been made public.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Databricks Acquires Electric to Bring Local Postgres Databases Inside AI Agents

Databricks is acquiring startup Electric for an undisclosed sum to integrate WebAssembly-based local Postgres databases directly into AI agent environments. Electric's key assets include PGlite, a lightweight Postgres that runs inside browsers or app sandboxes, and Electric Sync, a tool that keeps local data aligned with a central database. The acquisition addresses a core bottleneck in agentic AI: frequent round trips to remote servers that introduce latency and slow down autonomous decision-making. PGlite will handle an agent's immediate local state, while Databricks' recently launched Lakebase platform will serve as the central store for shared enterprise data. The move also has cost implications, as reducing remote database calls lowers network strain and cuts compute overhead from maintaining large numbers of active cloud connections.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How European SaaS Teams Can Build Safe, Idempotent Expired Session Cleanup

A technical architecture guide outlines how European SaaS platforms can reliably clean up expired user sessions without risking data errors or overloading worker systems. The recommended approach uses an authenticated HTTP trigger to enqueue cleanup batches rather than deleting sessions directly during a scheduled request. Four core invariants govern the design: a fixed expiry timestamp per run, deterministic batch keys, rate-limit-aware queue admission, and conditional deletion that checks session state at commit time. This last safeguard prevents a race condition where a refreshed session could be incorrectly deleted based on stale data. The architecture separates scheduler responsibility from application correctness, ensuring that delayed, duplicated, or retried triggers never corrupt the final set of active sessions.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer Builds Puerto Rican Mountain Kitchen Using Only HTML and CSS

A developer created 'Niebla', a one-page interactive concept kitchen inspired by Puerto Rico's mountain culture, built entirely with HTML and CSS — the only image file used is the favicon. Every visual element, including a steaming pot, rain on glass, fog, and floating roots, is rendered purely through code with no photographs, icon sets, or UI frameworks. The project centers on a sancocho dish whose ingredients users can interactively remove, with CSS custom properties dynamically adjusting steam speed and broth color in real time. Accessibility was a core priority, featuring bilingual Spanish and English support with proper ARIA live regions, and the developer removed a scroll-triggered reveal animation after discovering it hid content from headless browsers and screen readers. The project scored perfect 100s on Lighthouse for accessibility, best practices, and SEO, with zero external runtime assets loaded after the initial page paint.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

AI Workloads Drive HBM Demand, Squeezing DRAM Supply and Raising Prices in 2026

In 2026, AI training workloads are dominating global DRAM demand, with generative AI models requiring High-Bandwidth Memory technologies like HBM3E and HBM4 that far outperform traditional DDR5. Major manufacturers Samsung, SK Hynix, and Micron are redirecting production capacity toward HBM, reducing supply of conventional memory used in PCs, smartphones, and consumer electronics. Hyperscale operators such as Microsoft, Meta, and AWS are consuming memory at unprecedented volumes, with OpenAI alone reportedly securing 10 percent of global DRAM supply in late 2025. This supply squeeze has pushed DRAM prices up roughly 30 percent compared to 2024 levels, disproportionately hurting smaller hardware makers who lack the financial leverage of large cloud providers. Geopolitical tensions, including the US-China trade dispute and China's push to develop domestic HBM alternatives through firms like ChangXin Memory Technologies, risk fragmenting the global memory market into separate regional ecosystems.

0 comments Read more at DEV Community