Sign in Subscribe

Bob M

GLM-5.2 vs GPT-5.5: The Free Model That Out-Honests GPT

GLM-5.2 vs GPT-5.5: The Free Model That Out-Honests GPT

Key Takeaways: - GLM-5.2 admits uncertainty significantly more often than GPT-5.5, which tends to provide confident but incorrect answers. - GLM-5.2 costs less per million input tokens compared to GPT-5.5. It's cheaper on both input and output. - Per-task cost in production shows GLM-5.

Google's Researcher Left After a Short Stint. Then Another Notable Figure Quit Too.

Google's Researcher Left After a Short Stint. Then Another Notable Figure Quit Too.

Noam Shazeer announced he's heading to OpenAI. VP of engineering. Co-lead of Gemini. One of eight authors on the "Attention Is All You Need" paper. The one that invented the transformer architecture powering basically everything you're shipping right now. Shortly after, John Jumper followed.

The Benchmark That's Eating AI Alive

The Benchmark That's Eating AI Alive

Key Takeaways - SubQ achieved impressive results on a benchmark at a high token count, just barely ahead of another model. - On a different benchmark, SubQ had a notable production score, while the other model managed a significantly lower score. - Running the benchmark at a high token count

AI Scored High on Diagnosis. Doctors Scored Lower. Here's What That Actually Means

AI Scored High on Diagnosis. Doctors Scored Lower. Here's What That Actually Means

A recent study explored the use of ChatGPT through clinical vignettes. The AI scored high, while the median physician score was lower. Physicians using AI scored slightly higher, and the AI outperformed both. The difference was significant. That's the headline everyone quoted. Here's the part they

144 Mastra Packages Got Backdoored. Here's the Unfiltered Story.

144 Mastra Packages Got Backdoored. Here's the Unfiltered Story.

That's all it took. Recently, someone snuck malicious code into multiple @mastra npm packages. If you've got any Mastra stuff running anywhere in your stack, read this now. Not after your standup tomorrow. This isn't some hypothetical. Researchers were still seeing active connections on

Data Centers Will Use 12% of US Electricity by 2028. Here's Why That Number Is Worse Than It Looks.

Data Centers Will Use 12% of US Electricity by 2028. Here's Why That Number Is Worse Than It Looks.

Key Takeaways - The Department of Energy projects data centers will consume up to 12% of all U.S. electricity by 2028 — a figure utilities were not prepared for. - Frontier AI companies will need large data centers to train a single model. - North American colocation vacancy has fallen

Pentagon Deploys Grok AI for DOD Users

Pentagon Deploys Grok AI for DOD Users

TL;DR - Grok AI hits a significant number of DOD personnel through GenAI.mil. - The model shares rack space with Google's generative tools inside the same portal. - An "all legal uses" clause covers intel briefs, logistics, and paperwork without hard guardrails. - Budget