MillionOClock

I really hope someone from any of those companies (if possible all of them) would publish a very clear statement regarding the following question: If I build a commercial app that allows my users to connect using their OAuth token coming from their ChatGPT/Claude etc. account, do they allow me (and their users) to do this or not?

I totally understand that I should not reuse my own account to provide services to others, as direct API usage is the obvious choice here, but this is a different case.

I am currently developing something that would be the perfect fit for this OAuth based flow and I find it quite frustrating that in most cases I cannot find a clear answer to this question. I don't even know who I would be supposed to contact to get an answer or discuss this as an independent dev.

EDIT: Some answers to my comment have pointed out that the ToS of Anthropic were clear, I'm not saying they aren't if taken in a vacuum, yet in practice even after this being published some confusion remained online, in particular regarding wether OAuth token usage was still ok with the Agent SDK for personal usage. If it happens to be, that would lead to other questions I personally cannot find a clear answer to, hence my original statement. Also, I am very interested about the stance of other companies on this subject.

Maybe I am being overly cautious here but I want to be clear that this is just my personal opinion and me trying to understand what exactly is allowed or not. This is not some business or legal advice.

I just discovered this site and found it interesting, just a bit of feedback: AFAICT there is no way to filter between paywalled and non-paywalled websites, I wish there was one!

While there most likely is going to be some bias in the training of those kinds of models, we can also hope that transfer learning from other non-driving videos will at least help generate something close enough to the very real but unusual situations you are mentioning. We could imagine an LLM serving as some kind of fuzzer to create a large variety of prompts for the world model, which as we can see in the article seems pretty capable at generating fictive scenarios when asked to.

As always tho the devil lies in the details: is an LLM based generation pipeline good enough? What even is the definition of "good enough"? Even with good prompts will the world model output something sufficiently close to reality so that it can be used as a good virtual driving environment for further training / testing of autonomous cars? Or do the kind of limitations you mentioned still mean subtle but dangerous imprecisions will slip through and cause too poor data distribution to be a truly viable approach?

My personal feeling is that this we will land somewhere in between: I think approaches like this one will be very useful, but I also don't think the current state of AI models mean we can have something 100% reliable with this.

The question is: is 100% reliability a realistic goal? Human drivers are definitely not 100% reliable. If we come up with a solution 10x more reliable than the best human drivers, that maybe has some also some hard proof that it cannot have certain classes of catastrophic failure modes (probably with verified code based approaches that for instance guarantees that even if the NN output is invalid the car doesn't try to make moves out of a verifiably safe envelope) then I feel like the public and regulators would be much more inclined to authorize full autonomy.

I haven't read anything about this but I would also suppose long distance human intervention cannot be done for truly critical situations where you need a very quick reaction, whereas it would be more appropriate in situations where the car has stopped and is stuck not knowing what to do. Probably just stating the obvious here but indeed this seems like something very different from an RC car kind of situation.

Let's just hope the person you are responding to isn't Elon Musk!

Multiple config files of Xcode projects are not publicly documented as far as I remember and personally I have preferred to require my agents not to modify them out of fear it might break something and be hard to fix. I don't know how agentic programming will work in Xcode but I would expect it to do it using a safer approach, so that's also another case where it might have an advantage.

Your workflow looks very interesting especially the describe_ui part, are you already able to do this today?

Can this be used on iOS somehow? I am building a Swift app where this would be very useful but last time I checked I don't think it was possible.

On of the advantages of GitHub Copilot for me is that in terms of billing I find it very generous, depending on how you use it.

The thing is that on macOS at least, Codex does have the ability use an actual sandbox that I believe prevents certain write operations and network access.

Is this something that OpenAI explicitly approves per project? I have had a hard time understanding what their exact position is.

I don't know about Claude Code but in GitHub Copilot as far as I can tell the subagents are just always the same model as the main one you are using. They also need to be started manually by the main agent in many cases, whereas maybe the parent comment was referring about calling them more deterministically?

Then maybe the engine should be less about really simulating the 3D world and just trying best to preserve consistency, more about providing memory and saving context for consistency than truly simulating a lot besides higher level concerns (at which point we might wonder if it couldn't be directly part of the model somehow), but writing those lines I realize there would probably still be many edge cases exactly like what you are describing...

My theory, based on what I would see with non-thinking models, is that as soon as you start detailing something too much (ie: not just "speak in the style of X" but more like "speak in the style of X with [a list of adjectives detailing the style of X]" they would loose creativity, would not fit the style very well anymore etc. I don't know how things have evolved with new training techniques etc. but I suspected that overthinking their tasks by detailing too much what they have to do can lower quality in some models for creative tasks.

An hybrid approach could maybe work, have a more or less standard game engine for coherence and use this kind of generative AI more or less as a short term rendering and physics sim engine.

I love AI but I also hope it will paradoxically make people realize the value of real life experiences and human relationships.

Recently I was thinking about how some (expensive) customer electronics like the Mac Studio can run pretty powerful open source models with a pretty efficient power consumption, that could pretty easily run on private renewable energy, and that are on most (all?) fronts much more powerful than the original ChatGPT especially if connected to a good knowledge base. Meaning that aside from very extreme scenarios I think it is safe to say that there will always be a way not to go back to how we used to code, as long as we can offer the correct hardware and energy. Of course personally I think we will never need to go to such extreme ends... despite knowing of people who seem to seriously think developed countries heavily run out of electricity one day, which, while I reckon there might be tensions, seems like a laughable idea IMHO.

I very recently (~ 1 week ago) subscribed to the Pro plan and was indeed surprised by how fast I reached my quota compared to say Codex with similar subscription tier. The UX is generally really cool with Claude Code, which left me with a bit of a bittersweet feeling of not even being able to truly explore all the possibilities since after just making basic planning and code changes I am already out of quota for experimenting with various ways of using subagents, testing background stuff etc.

Maybe I am missing something but the last few times I tested VMs it seemed to end up never shrinking in RAM size once it had grown, is this a real issue and if so is there any improvement coming on that front?

I’ll use this post as an occasion to ask a question I can’t find an answer to anywhere as someone developing a closed-source commercial agentic AI builder app: does either GitHub Copilot or OpenAI codex teams plan to also expand this quota usage support to all tools provided they respect certain rules, even if non open-source and/or commercial? Up to now I was planning to just enable people to input their own API key but this kind of integration would be amazing for me. Would really love some clarification about this if anyone from those teams reads this, or if anyone knows who to potentially contact to find an answer.

Am I understanding it correctly, based on these tweets [1][2], that both Codex and Copilot teams or at least team members mentioned potentially letting people make use of their quotas in third party tools?

I really would like further clarification on those points as I would be pretty interested for a product I'm building if it was indeed made possible.

[1] https://x.com/jaredpalmer/status/2009844004221833625

[2] https://x.com/thsottiaux/status/2009714843587342393

Wow, exactly the same issue for me, and for two different accounts of mine!

I also feel like an heavily multimodal model could be very nice for this: allow multiple images from various angles, optionally some true depth data even if imperfect (like what a basic phone LIDAR would output), why not even photos of the same place even if it comes from other sources at other times (just to gather more data), and based on that generate a 3D scene you can explore, using generative AI for filling with plausible content what is missing.

It should be noted that OpenAI now has a specific compaction API which returns opaque encrypted items. This is AFAICT different from deciding when to compact, and many open source tools should indeed be inspectable to that regard.

What are your favorite features? I recently downloaded it and also use Codex CLI and GitHub Copilot in VS Code but I don't really know what specific features it has others might not have.

It’s unclear to me wether that would give some access to a token quota or if it would just be like any other « Sign in with … ». In all cases I am currently developing an app that would greatly benefit from letting my users connect to their ChatGPT account and use some token quota.

The issue I see is that for certain apps, such as one I am currently working on and hope to publish soon on iOS, is that they do require a lot of maintenance once published even if there were no server costs. Given the amount of work I already put in it and how much more will be necessary even just to keep the app correctly running in the future, I don't really see what other monetization approach would make sense for me. Actually, I would even argue that selling an app without a subscription might (sometimes) be setting wrong or blurry expectations: if a user accepts to pay today a single time, how long are they expecting updates for? Will it only be basic bug fixes or also major new features? With a subscription, I feel like at least if they are unhappy with my app, they won't really have lost anything and can just unsubscribe, since they had basically accepted, IMO, that the money they put in my app each period of time is only for the service and potential updates in that small period of time and not future changes.

What is the maximum context size in the web UI?

Same for me here

I frequently have exactly the same double speed issue on YouTube, nice to know I'm not the only one

I wonder the same too, also is Gemini Pro supposed to beat GPT-4?