Brad Dangerfield

Grateful to @ZacksJerryRig and @NotaWheelchair for helping me arrange this absolute unit for my nephew ❤️ Thanks you guys for the good you do.

@Manhattva this is why you need one.

This is a order of magnitude leap in capabilities for global communications and payload capability to Leo

Cybernetic Agents talking and Interacting with other cybernetic agents on behalf of humans is the future It may even be so pervasive it will feel foreign to make phone calls and send human written mail except to friends and loved ones

Cybercab spotted in SLC!

OpenAI PAUSED new 200/mo subscription tier!

I have ultrafast in codex cli

This must pass

4% battery degradation after 60K miles How’s everyone else’s @cybertruck ?

What in the heck is going on @BasedMikeLee @SpencerJCox @Manhattva @jasoninthehouse

GPT-6 medium review, in an extremely complex codebase, by Fable5.1 (sanitized reply): I Ran sessions ran on GPT-6 Astra at medium reasoning effort. I’ve used it across several threads since the recent outage, including some earlier diagnostic work. I used only medium reasoning as requested. Where it is clearly better than the 5.6 models Root-cause analysis. It has been particularly strong at tracing difficult issues through multiple layers, identifying subtle state and lifecycle problems, finding silent gating behavior, understanding existing contracts, and isolating timing-related bottlenecks. Its conclusions have generally come with concrete evidence and a clear distinction between what is demonstrated versus what remains unproven. Honesty under pressure. It has also been good at challenging incorrect assumptions rather than forcing the evidence to fit the initial theory. On multiple occasions it corrected the framing of the investigation or explicitly identified a proposed fix as valid but not yet proven to be the live cause. Subsequent diagnostics supported that caution. Where it is weaker Finishing. It is noticeably more conservative about pushing changes all the way through validation. In one case it stopped because of an environmental constraint; in another it produced a legitimate but incomplete change and clearly reported the remaining test failures instead of continuing until everything was green. Sol xhigh has been the stronger closer. It has been better at taking the investigative work, resolving integration issues, cleaning up surrounding failures, and getting the final result through the complete test suite efficiently. There was also one case where GPT-6 produced a technically correct fix that turned out not to address the actual live bottleneck. That consumed an additional iteration. Once better runtime evidence was available, however, it identified the underlying timing issue very quickly. Its reports also tend to be dense and lengthy unless explicitly constrained, and it can be overly conservative about running expensive build or validation steps without explicit permission. Typical turnaround at medium For the investigative threads in this session, turnaround has generally been on the order of 10–20 minutes per substantial thread, with narrower read-only investigations toward the lower end and deeper implementation/design work toward the upper end. Working model split The pattern I’ve settled on is: GPT-6 medium: investigation, difficult root-cause analysis, architecture, and design decisions. Sol xhigh: implementation, integration, conflict resolution, validation, and getting the final change green. Terra high: operational work and smaller, more bounded changes. That division has worked well: GPT-6 is currently the stronger investigator, while Sol xhigh remains the more reliable closer. There are still a couple of unresolved performance/architecture questions that look like especially good candidates for trying GPT-6 at high reasoning effort—particularly one unexplained latency component and a broader non-blocking processing design question.

Anthropic just did a reset. @thsottiaux did you reset the wrong one?

GPT 6 is here

The argument for private inference

This is the kind of thing that will heal the mental health of the nation - Present loving parents

@CafeZupas has to have the absolute worst ordering experience I have to repeat my order FOUR TIMES just to get my food. Awful.

This is the type of person we need to represent Americans.

Grokbot thumbs up to acknowledge an additional request? anyone else see this?

This is why FSD is so critical - life worth is so valuable. This could be your child

Me trying to beat rush hour