I tried codex on windows for first time today (typically use cli on osx/wsl linux). Codex app on windows is unusable (relative to my alternatives):
- if i make a new chat, it takes 10-20seconds(!!) to set up environment, whatever that means. I wasn't even using it for coding. It was pointed at empty dir.
- i tried browser usage for first time. Asked it to sign into website. It said it did, and I should log in. But no browser displayed.
- After pointing this out, it showed the browser sidecar. Typed in password. submit. Nothing happens.
- Installed chrome openai extension. that worked better. Asked it to save a few files to my project as pdfs. It tried, but can't control print modal. So that failed too.
- Overall, total failure.
This dialog doesn't take into account parking. I disengage ~every drive, during parking. At home, I need to back up into garage. At work, I need to go to specific spot in parking garage.
At gym, I let it park itself once. It did a horrible job, not evenly between lines.
When I route to school to drop kids off, sometimes it picks parking lot A (correct), sometimes lot B (not for dropoff).
Either remove the dialog for parking, or provide a better way for me to instruct car about exactly where to park (with fallbacks if its full). And be sure parking is dialed in (it's not right now, for cybertruck at least).
1. Auto-improving a codebase along various dimensions
2. Spec driven development of some kind
3. Both of these feed some kind of queue-based system where multi-agents pull tasks from
4. Some kind of configurable pipeline/workflow idea (1. code it. 2. localized review. 3. polish PR 4. big picture review (conformance to spec, bigger picture, overall architecture), 5. back to earlier step or ship it)
5. configurable/transparent system to manage learned preferences and learnings from mistakes
@Investanswers He is bitter. Also, it’s a real issue, and has happened to me until I got car serviced (condensation in camera housing). His error is the same as yours (X happens for me, so therefore is the only truth).
I'm trying to build my own tooling against gpt-5-codex, but running into clear OAI bugs where function calls result in an endless stream of response.function_call_arguments.delta events. Often whitespace (eg, "\n\n\r"), but today i even saw an endless loop of "functions.multi_tool_use.parallel to= functions.multi_tool_use.parallel to= ... https://t.co/RNWgyDb536_file to=...". Wondering if you can help? I'd be happy to devote any amount of my time to get to the bottom of this
Same. On my first v14 drive, 2 hard breaks at yellows (humans, and v13, would go thru). Then hesitancy making unprotected left left me in the middle of road when it decided against it. I had to take over and back up. Neither are safety issues per se, but it’s a step back.
OTOH, it successfully parked in lot at destination and escaped a parking garage, which is cool.
@sama This is the wrong UX. The UX I want is more like asking questions in a comment on a google docs (directionally speaking). Opening a new page takes me out of flow, and I can't see the whole picture.
@bradsferguson The Waymo soln will have proven superior if it is first to market for robotaxi (it is) and if early market share lead is durable (remains to be seen, but I’m skeptical. But this is certainly possible).
@shai_wininger@Lemonade_Inc I'm curious -- does this mean you're enforcing all code is written by AI? Or simply that AI is used "a lot" by all the engineers? Is there room for hand coding or debugging some things the AI isn't able to solve? If the latter, how do you know if you're using AI "enough"?
@jonbbc I’m also confused about this insanely glowing feedback about 12.5. It’s better but still has multiple interventions per drive, some safety critical. Biggest thing that got better for me is stop sign behavior and confidence. But still so many issues
I got it last night. It’s much better and I like it. Nonetheless, 2 safety critical interventions (one failure to stop for peds crossing street at stop walk, one inexplicable “must take over now” despite nothing unusual. 2 failures to be in correct lane. 1 random slowdown for no reason. 1 failure to go despite all clear. And 1 driving 25 in a 40 with people honking at me from behind. This is just a few trips, maybe 10-20 miles. Progress? Yes. Holy grail? Not close
Stop lights are handed well. Stop signs are quite bad. Residential driving too slow. Centering in lane is worse, including when turning. Overall 12.3 is solid for me, but it’s clear there’s still a long way to go. Bulls too excited imo. But I’m keeping a close eye, still very good progress.
@alojoh@TSLAFanMtl I follow both of you. You’re both good analysts. AJ, you unnecessarily attack nearly everyone. The height of arrogance. You don’t just attack ideas. You attack people. Over and over.
@EmmetPeppers@garyblack00 So you’re saying Elon is purposefully hurting Tesla as a negotiation tactic? I know you didn’t explicitly say that, but my naive reading was that’s the implication
@mattforsberg@Tesla Do you happen to know if this product is compatible with Enphase inverters? I know PW3 has built in inverters, but I already have solar. I’d prefer PW 3 due to LFP.
I understand this reasoning. At the same time, as a customer of yours, it’s useful to have a metric that monitors the success of your ideas. As you know, investing requires being both directionally correct and timely. It’s not easy to look back over your work to assess both.
Perhaps a simplified model portfolio would work. It wouldn’t need to include all your trades, or frankly, even be your actual trades. Only that it’s an immutable log that can be assessed.
That being said, I understand why you wouldn’t want this. Not the least of which, ideas like yours usually don’t ALWAYs perform well, which would make it awkward to sell more subscriptions during such periods 😂 Your thought on not providing financial advice is understandable but not compelling to me. All my decisions are my own, period, regardless of what analysts say.