OpenAI just revealed the first results from Jalapeño, its first custom AI inference chip, and the gains are huge. Across big models including DeepSeek R1 670B and Kimi K2.5 1T, it delivers 1.5-1.9x more AI work per watt, 1.7–3.6x lower end to end latency, and up to 4.1x higher performance for highly interactive workloads versus comparison systems. AI helped design the chip, enabling OpenAI to go from initial design to tapeout in only 9 months. AI generated implementations for selected workloads were already 1.5–1.8x faster than human expert written versions. AI designs better chips → better chips run AI faster and cheaper → stronger AI designs even better chips → repeat

Teams leaving the OpenAI API aren't mad at OpenAI. They're tired of betting everything on one provider's pricing and rate limits. Claude and Gemini are legit peers now. GLM 5.2 and DeepSeek R1 are closing fast. Gateways run all of them with zero code changes. Full breakdown:

Incoming UC Berkeley prof. @sayashk reveals AI reliability is improving 4–10X slower than capability and why 90% accuracy can still be catastrophic: "We came up with a list of metrics that other industries had used, industries like aviation and nuclear safety and so on. We tried to port these metrics to the AI agent space." "Reliability improving much more slowly than capability. Based on these early results, it was between four and 10 times as slow as capability improvements have been in terms of accuracy numbers." "That's why AI's effect on the job market has been augmentation not automation. You can't accept an error rate of even 1% or 5%." "Rabbit R1 and Humane Tech Pin were personal agent assistants. If your product orders your DoorDash food to the correct address 90% of the times, it's a catastrophic product failure because 10% of the times your users are mad at you." @UCBerkeley

Tesla raised prices on the Cybertruck 6-7% (Standard and Premium) likely to reduces loss per sold vehicle after the collapse in volume provides very little fixed cost absorption making profitability nearly impossible. Currently Tesla sells Cybertrucks at only around 15K units annually. That's not even 10% of the targeted 250K annually. "I think we’ll end up with roughly a 0.25 million Cybertrucks a year, I think we’ll probably reach it sometime in 2025." (Elon, October 2023) At below 10% target achievement, to call the Cybertruck a commercial failure would be objectively an understatement. The problem is that in automotive significant overestimation causes a major cost problem since production is designed for a certain volume. Tesla's production system is completely overbuilt for the tiny volume relative to designed production. This excess infrastructure incurs excessive accounting charges. Tesla is lucky however that it can access capital at almost zero cost. A good example is the Rivian R1 platform. This platform and production system too was overbuilt. Forcing Rivian to completely redesign the R1 platform after only 2 years and tearing the old production system down and replacing it with a new one. This obviously resulted in a large one-time write-off but similar to Tesla, Rivian too had access for a time to incredibly cheap capital (not anymore). The more capital a company has to more mistakes it gets to make and it can still come out on top.