Techbible
Share
28/05/2026
Anthropic just dropped Claude Opus 4.8. Same price as 4.7. Agentic coding jumped from 64% to 69%. Multidisciplinary reasoning up to 58%. Fast mode 2.5x quicker. 4x less likely to let coding errors slide. Alignment now matches their Mythos preview. Honesty is the new benchmark.
28/05/2026
Copenhagen & SF may 26 ! In no particular order may life recap š«¶
The most valuable company on earth just got out-spent by its own engineers. š
Everyoneās racing to adopt AI. Almost nobody can tell you what itās actually returning. Thatās the gap ELI closes.
Weāre building payroll for your AI workforce. Every agent, every tool, every token, tracked and tied to the value it creates. So you never have to cancel your best performer just because the invoice got scary.
Your AI bill is growing whether youāre watching it or not. š
Comment āELIā and Iāll send you what weāre building.
27/05/2026
5 papers I read this week so you donāt have to š§
ā Multi-agent AI systems fail because of topology, not weights and scaling makes it worse ā Your RLHF is killing diversity. Vector Policy Optimization fixes it ā DeepMindās Lean agent just cracked 9 open ErdÅs problems autonomously ā Personal agents need to live on your device, not the cloud ā Anthropicās Mythos model writes zero-day exploits no other model can touch
21/05/2026
Paul ErdÅs posed a geometry problem in 1946 and offered cash to anyone who could crack it.
80 years. Hundreds of mathematicians. No one got it.
An OpenAI model just did. Autonomously. From a written problem statement. No human in the loop.
The āAI is just autocompleteā crowd needs a new bit.
GitHub confirmed attackers accessed around 3,800 internal repositories after an employee device was compromised through a poisoned VS Code extension.
The breach reportedly stayed limited to GitHubās internal systems, with no evidence of customer repo exposure so far.
Big reminder that developer tools and extensions are now a major supply chain attack surface. One compromised plugin is enough to open the door.
19/05/2026
š„
One of the original minds behind OpenAI is stepping into a new chapter focused on large-scale model training and foundational AI research.
Karpathy says heās excited to return to hands-on technical work after spending time focused on AI education and independent projects.
Another major talent shift in the AI race. š
Crazy ideas I ask my friends about .overdrive from
Which database should power your AI agent? š¢ SQL (Postgres/MySQL): Reads every row. Accurate, structured, slow. Great for analytics, painful for real-time agents. š MongoDB: Flexible schema, but searching unstructured docs at scale = latency. Good for storage, mid for retrieval. ā” Redis: Lives in RAM. 0.2ms response. Vector search + session memory + caching in one. Built for AI speed. Latency is the new accuracy.
ššš .ai .work_
Click here to claim your Sponsored Listing.
