It's impressive that you didn't reach for Claude Code first :)
Would be good to keep track of how long it takes you to complete, so you can compare it with how long it takes Opus and then Fable to accomplish the same thing.
> It's impressive that you didn't reach for Claude Code first :)
I should have. All I have right now is a free Cursor account and Github Copilot. I've got to start spending money on tokens and get up to date. That article today by Carmack chewing out game devs for hand coding was striking, considering who he is and how much low level stuff he has done.
I'm doing 3D rendering in Rust. (Think three.js, but in Rust.) This is well outside the mainstream, so there's not much on the web to populate training sets. A year ago, AI coding tools could not cope with this area at all. Today, maybe.
I have no idea how good Claude is at understanding breaking changes to an API. I expect it would get simple changes, such as structure field renamings, without any trouble. But the gyrations required to handle write only access to GPU memory in Rust might not appear in the training data.
To add to this, merely using the thumbs up/down button in a chat could share your entire conversation with them for model training.
From their docs[1] (archive copy is at [2]):
> You can opt out of training through our privacy portal by clicking on “do not train on my content.” To turn off training for your ChatGPT conversations and Codex tasks, follow the instructions in our Data Controls FAQ. Once you opt out, new conversations will not be used to train our models.
> For a linked teen account, a parent or guardian may manage whether conversations can be used to improve our models through Parental controls.
> Even if you have opted out of training, you can still choose to provide feedback to us about your interactions with our products (for instance, by selecting thumbs up or thumbs down on a model response). If you choose to provide feedback, the entire conversation associated with that feedback may be used to train our models.
I mean, how else would those buttons work? It's explicitly feedback data. And "this is good" or "this is bad" is empty if divorced from what "this" actually is.
If the buttons are incompatible with the absence of the feature, I'd expect the buttons not to exist when the feature is disabled. Anything else seems like a straight up footgun. I guess it'd also be acceptable to pop up a scary warning box asking "are you sure?"
> If the buttons are incompatible with the absence of the feature, I'd expect the buttons not to exist when the feature is disabled. Anything else seems like a straight up footgun.
It's called a "dark pattern." They want you to shoot yourself in the foot, so they'll do their best to aim your gun at your foot and put your finger on the trigger. And then when you do, because you don't have perfect understanding or execution, they'll say "your fault!"
It could go into personalization / memory. Or they could be A/B testing some system prompt tuning and consider the thumbs up / thumbs down as statistical feedback on the particular flags that are enabled for your account.
This comic was made a loooooonnnnnnggggg time ago (back when I worked at Facebook). It was actually a great place to work then, but that company no longer exists in any meaningful sense.
No, you're off by one - likely because the OpenAI models punch above their weight in those comparison, hence my original message. You've shifted the comparisons to favor Anthropic.
Fable/Mythos are much larger than Sol. They match to Astra which is supposedly at least 10T. Astra is already publicly confirmed as a new family.
Opus matches to Sol. Sonnet to Terra. Haiku to Luna.
Anthropic is able to compete at the frontier high-end by launching massively large expensive models. But their inability to compete on small models belies their efficiency aspirations across the stack.
Spot on wrt CoT. I have thinkingSummaries enabled and I find it eminently readable compared to the prose in Claude's replies.
In fact, whenever Claude disobeys me, I usually first skim the CoT to figure out if my original instruction was ambigous given the context. I usually come away with a better understanding of how to frame my prompt to be less ambiguous or just force myself to be more explicit when prompting.
Regarding diosbedience, usually this is either due to a blanket instruction from me during an earlier turn in the same session, an explicit instruction in its system prompt or it being just eager to bring a task to completion.
Chain of thought does not exist in the output of Claude, they disabled true thinking due to distillation risk. What you see when thinking summaries are enabled are just that, summaries of thinking into Claude-isms, therefore you cannot make any inferences on what the model is doing unless you literally work at Anthropic and can see the true thinking traces.
Sure, but that doesn't tell you about how the CoT is phrased when the agent is its own target audience, which is the interesting thing under discussion here.
They want to be able to zoom in and zoom out as needed while analyzing aggregated user requests to better understand the different kinds of ways (well-resourced) actors use to distill their most capable models.
> "The data will help us defend against complex and novel attacks (including new jailbreaks and attacks that operate across many requests) as well as help us identify and reduce false positives."
> "Some attacks only become visible across multiple requests. Best-of-N jailbreaking, for example, sends hundreds of slight variations of a prompt in the hope that one will work. Larger patterns of misuse, such as state-sponsored espionage or data extortion campaigns, only surface when our safeguards classifiers can zoom out across many requests. Detecting these threats requires temporarily retaining prompts and outputs so they can be analyzed together, rather than one at a time."
https://x.com/RampLabs/status/2042672773747589588
reply