Robert Dylina

@sama Any chance for a good open source conversational model from @OpenAI at some point?

pretty excited for voice models to get great its interesting to watch how people are already starting to change the way they interface with AI

@sindresorhus Ask codex to check why it is running slow, ask it to pay particular attention to your session file sizes. You may be experiencing the same problem I am.

Oh how I wish the Codex app was native. It's so slow lately... Why can they not just make Codex rewrite it in Swift?

@Obserfessor @ItsTheDaybreak @thsottiaux That's interesting, I wonder if it interprets that language to be more robust in its output.

@rdylina @ItsTheDaybreak @thsottiaux I go further and ask ChatGPT (or other LLM) to generate a context file, and then I upload that context file.

What are we obviously not getting right with Codex?

@TumbleweedChase @MattLoszak Several people have pointed this out. I guess my question has more so become, "is it more cost effective to do low temperature despite the efficiency losses". So far no one has been able to do the math for me. I suspect pricing isn't very public on equipment like this.

@rdylina @MattLoszak Because that is horribly inefficient. High temp = more efficient. Nuclear reactors get away with it as the fuel is so cheap.

Turbines for nuclear are easier to buy than turbines for gas plants. We bought our first one a few months ago, and it's arriving later this year. Nuclear plants use *steam* turbines, which operate at lower temperatures than combusting gas. Turbine blades are nowhere near as big a bottleneck for nuclear & steam. A modern gas turbine inlet runs at 1,500 C. Light water reactor steam turbine inlets run at 285 C. Aalo's steam turbines will run closer to 400 C. At 285 - 400 C, you can use more common metal alloys and manufacturing techniques. No single crystals, no vacuum furnaces, no 90-week grow times. 3 companies on Earth can cast a hot-section gas blade, while over 10 companies can forge a steam turbine rotor, for the smaller turbines used by smaller nuclear reactors like Aalo-1. Steam turbines (esp <= 50 MWe) are a more mature commodity. There's more to this story though, and a fascinating history... going to post a follow up soon.

@thsottiaux Also, session pruning. Sessions currently grow indefinitely until they're so big they choke the app to a crawl. Long running heavy sessions hit 1gb+ in a few days.

@bcchen82 @thsottiaux This has never happened to me once. Weird

@thsottiaux I am getting 50% or even more error rate of compacting, stream disconnect before compact succeed.

@harsette @thsottiaux I'm having good luck having it pass image gen the UI specs and then having it give me variations. Once I pick one I tell it to replicate the reference image until it has matched as much as possible.

@thsottiaux FRONT END DESIGN TIBO 😭😭😭😭 It's unbearable, we shouldn't have to jump in circles to get a decent looking ui heyll nah

@ItsTheDaybreak @thsottiaux In the meantime I ask chatgpt to give me a prompt to carry to another agent with all the relevant information. Does a good job for now but good idea.

@thsottiaux You need to make Codex link up with ChatGPT in some way. I have had many meaningful conversations on ChatGPT, but I currently cannot access those memories or see them displayed within Codex.

@techn0eacc @thsottiaux In the meantime you could solve this with a couple of instructions in your global prompt and if you're feeling frisky you could integrate @mattpocockuk grill me skill as part of those instructions.

@thsottiaux I wish Codex would step in with questions when the user wasn't specific enough about their request. Like what'd be the purpose of the app user wants to create.

@Zenoware @scaling01 @thsottiaux That computes

@scaling01 @thsottiaux /goal has obliterated my weekly quota in a span of two days lol

@scaling01 @thsottiaux

@thsottiaux the rate limits constantly running into them need more resets

@thsottiaux Running multiple agents with computer use is very difficult. Would be easier if we could support multi-desktop sessions so each agent could have its own dedicated screen or VM.

I think merge requests is a better name then pull requests

@MollySOShea The display of agility in the background is amazing

More body positions than atoms in the universe? "The robot’s made up of ~40 motors. Every motor can spin 360 degrees, all the way around. So mathematically, how many states it could be in? Body positions, is 360 to the power of 40." "The difficulty here is, how do you control it? You can’t write code to make this work." Brett Adcock (@adcock_brett) Founder & CEO of Figure (@Figure_robot)

Massive sessions is an issue for me. But I can't just discard and handoff to a fresh chat. Since codex is an everything app now, I have persistent very long running sessions with automations as part of workflows. I can't manually create a new thread and hand off and redo the automation every few days. In reality, codex sessions need to become self-pruning so they don't grow indefinitely. I have some that are over 1gb when it realistically only needs replay for like the last 100mb AT MOST.

@meta_alchemist Make this into a skill, would be useful!

Codex's app has been super slow for me lately. at first, I thought the problem was Codex itself. It wasn’t. After cleaning things up properly, Codex felt roughly 10X faster. 0 slowness. Before this, I had 8GB of logs built up, and it slowed things down like crazy. Here’s the 15-point cleanup system, which worked perfectly for me. It won't delete anything. Copy paste these 15 bullet points when your Codex starts to slow down: > it will inspect things first > back up & archive important files > and make your Codex blazing fast again. 15 ITEMS TO KEEP CODEX FAST 1. Check what is actually taking space. Inspect sessions, archived sessions, worktrees, archived worktrees, logs, config, and the local state database. 2. Back up the important files first. Back up config, global state, session index, state database, memories, skills, plugins, and automations before changing anything. 3. Check if Codex is open. If Codex is running, only inspect. Apply cleanup after closing it so the local database is not being touched from two places. 4. Find the giant active chats. Look for the biggest active session files. These are often old conversations that are still treated as active history. 5. Archive old non-pinned chats. Move chats older than 7-10 days into archived sessions, unless they are pinned or clearly still current. 6. Keep only recent work active. Your sidebar/history should not be carrying weeks or months old execution threads. 7. Use handoff docs instead of massive chats. If an old thread matters, turn it into a handoff doc, archive the thread, and resume in a fresh chat from the doc. 8. Normalize weird paths. On Windows, clean up path mismatches like normal C:\... paths vs extended \\?\C:\... paths. 9. Prune dead config projects. Remove project paths from config that no longer exist or point to temporary folders. 10. Move stale worktrees. Don’t keep old Codex worktrees in the hot worktrees folder. Archive them instead of deleting them. 11. Rotate large logs. Move oversized old logs into an archive folder so Codex can recreate fresh ones. 12. Check heavy background processes. Look at Node/dev-server processes. Don’t auto-kill them, but close the ones you don’t need. 13. Verify the cleanup. Afterward, confirm config still parses, the database opens, active session size dropped, archived sessions increased, and no bad paths remain. 14. Turn this into a weekly script. The cleanup should not be a dramatic one-time rescue mission. Make it repeatable. 15. Make it boring. Weekly maintenance should back up first, archive old sessions, normalize paths, prune config, move stale worktrees, rotate logs, and give you a report. The biggest lesson for me: giant chats should not become permanent memory. Chats are for execution. Handoff docs are for memory. Archives are for history. Fresh threads are for speed. P.S. Before doing all this, make comprehensive handoff documents for each active chat, too, with prompts prepared for each to reactivate them after. This will start new chats from the exact places you left off, but at blazing-fast speed. Like this, things simply work perfectly. I even told my Codex to automate these weekly, and it has set it up for every Sunday. Save this for when you will need it, as Codex app does get heavy as you use it more, especially if you are using many terminals and long sessions a lot.

I may have found a potential bug/issue? Apparently the session JSONL files never get trimmed. Ever. So if you have a large feature you've been working on for a week, the session file can grow enormous until it chokes codex to death. Or if you have a long running automation, same problem. I have a couple session files that got over 1gb and is causing severe lag across the entire application. Seems like there should be a hard cap before the head on these files starts getting trimmed.

It's also become super good after all the @steipete cooking over the last week or so. I've had the best experience with GPT-5.5 so far, it's night and day compared to two weeks ago.

@GhostOfFrederic @gbponz @MattLoszak That's an unhelpful engagement, but if you read the entire thread there is a great discussion explaining the how and why. Learned a lot along the way about advanced thermodynamics that wasn't covered in my electrical engineering classes once upon a time.

@rdylina @gbponz @MattLoszak You need to learn some thermodynamics.

@thechaosledger @MattLoszak That sounds great. Makes me wonder whatever happened to some of those groups that were working on trying to convert heat directly into electricity using carbon nanotubes years ago.

@rdylina @MattLoszak Because you're using the waste heat, you don't add any extra fuel. Takes cycle efficiency from ~35% for just the GT to ~60% for closed cycle

@Exogynous @MattLoszak Makes sense. Great insight

It all depends on how much you lose by not being able to scale and just how expensive the turbines get. With gas turbines you are hostage to the cost of gas. With atomics your costs tend to be almost all up front and then minor maintenance. If you are using gas to heat steam and only getting 30% back instead of 60%, that is going to double the cost of your power. I don't see doubling your production cost as survivable in the current environment. For Aalo it makes sense. Nukes run continuously at around 98% output and your costs are capped. But it is interesting to note that Aalo originally proposed sCO2 as the working fluid and high performance turbines running in the 1,200c to 1,400c range. This would allow them to get around 65% energy back without the rankine steam bottoming cycle. I suspect turbine availability has something to do with it but also starting smaller and more controlled is wise in a new product.

@Celynx_Mats @MattLoszak Security like volatility? Will that matter as much with the newer generation reactors that are designed to be impossible to start a runaway reaction?

@rdylina @MattLoszak Generally speaking you get a better deal with higher P and T at the inlet (within reasons). The low pressure end of the steam turbine is very costly due to the long blades (due to higher volume). But most nukes went without superheated steam due to security concerns.