ActivePattern
ActivePattern on Hacker News
Because a math proof is generally de-composable into smaller constituent arguments that can be independently evaluated for correctness. So you can audit a sample of these arguments for correctness without reviewing everything. You would then verify that the arguments being true implies the conclusion.
But the article clearly describes how this SCM is just instrumental in their goal to build and serve better models. It's not some unrelated product they hope to sell.
Post-training of LLMs often destroys calibration. The objective is no longer to predict the next token, so it doesn't matter as much whether the probability you assign to next-token "A" is correct, as long as the overall answer is good.
So it's not too surprising that the most heavily post-trained frontier models are actually pretty bad at outputting calibrated decisions.
The biggest upside to using a zero-shot classification model is that it's actually zero-shot, i.e., you don't need to construct task-specific training sets.
If it needs to be fine-tuned, then why not fine-tune smaller and cheaper models that are only as big as the task requires?
That's right. There's only so much optimization that you can make to a transformer-based model and any tricks that Jev is employing, any open-source LLM can also employ.
It's just marketing. More specifically, it's an answer for why their model can't answer questions that require reasoning.
"Reply with just the letter A, B, or C."
There, I fixed your problem.
It would be great to see benchmarks for Jev that demonstrate the value of calibrated uncertainty.
For example, one could set a confidence threshold over which we trust the model decision, and otherwise reject. This provides a lever to trade-off accuracy and automation %.
Then we can ask questions like "What % of decisions can we automate to achieve 90% accuracy"?
100% agree. The key issue is that users think they are getting results from a specific model configuration and they are clearly not, which is fraudulent.
If OpenRouter wants to succeed as a business, they need to be auditing the providers they connect to (i.e. benchmarking) and removing fraudulent ones from their service.
Yeah, the README seems like AI slop to me. That always makes me lose interest in a project because it seems so low-effort.
How are you supporting the claim that GPT 5.6 is "far more token efficient" than Opus 5? Tokens equal, output is cheaper for Opus 5 ($25/1M) than GPT-5.6-Sol ($30/1M), and it seems to outperform slightly on agentic coding benchmarks.
I think you're being overly cynical here. First, I don't see any claim that is the world's best model for agentic coding. Second, it is absolutely the best model in terms of coding performance vs. dollar, and it's raw performance seems very close to the frontier.
It's interesting that it cares about formatting choices like bullet points.
On the one hand, it makes the system more brittle and arguably overfit, since it's not making a decision on the content of the writing itself. Most people would say that AI writing is still AI writing even if bullet points are removed, and vice versa.
On the other hand, we know that AI writing does has a very specific formatting signature (e.g. em-dashes, bullet points) so it seems unwise to completely ignore it.
You would be surprised how much of a difference the model makes for certain niche tasks.
For my use case, `gemini-3.1-flash-lite` is ~20% higher accuracy than the next best model of comparable cost (considering both proprietary and open-weight alternatives)
A lot of people in this thread think they're being clever by pointing out that a bookstore can't sell "banned" books. But it's common for bookstores and libraries to feature titles that have been banned in some jurisdictions. It's a small way to push back against censorship and promote freedom of information and critical thinking.
It took a few minutes to get the hang of, but it does work very nicely! Being able to adjust length without any re-tying is a great feature
Seeing this makes me wonder if Grok uses Claude conversations for training.
It's otherwise kind of surprising that they both converge on very similar phrases (e.g. "API integration is kicking my ass") that aren't anywhere in the prompt.
Automatic coding systems have way too much economic value to be considered a "fad". I don't think you need to be Nostradamus to predict that we're never going back to manual coding. Sure, the systems will evolve and improve, but they're certainly not going anywhere.
You're describing "modularity" or "loose coupling" in code. But it rarely implies you can just delete files or directory. It usually just means that a change in one component requires minimal changes to other components -- i.e. the diff is kept small.
The win is in how many weights you process per instruction and how much data you load.
So it's not that individual ops are faster — it's that the packed representation lets each instruction do more useful work, and you're moving far less data from memory to do it.
I still don't understand is why they don't even make an attempt to apply overlayers, when (as the author notes) there is ample secondary evidence that it would be present. It's not like there isn't already some element of inference and "filling in the blanks" when reconstructing how something was painted from the scant traces of paint that survived.
I assume you didn't read the article, since that's their exact point...
"Since underlayers are generally the only element of which traces survive, such doctrines lead to all-underlayer reconstructions, with the overlayers that were obviously originally present excluded for lack of evidence."
> I can only say being against this is either it’s self-interest or not able to grasp it.
So we're just waving away the carbon cost, centralization of power, privacy fallout, fraud amplification, and the erosion of trust in information? These are enormous society-level effects (and there are many more to list).
Dismissing AI criticism as simply ignorance says more about your own.
I've never heard the caveat that it can't be attributable to misinformation in the pre-training corpus. For frontier models, we don't even have access to the enormous training corpus, so we would have no way of verifying whether or not it is regurgitating some misinformation that it had seen there or whether it is inventing something out of whole cloth.
Yes, as is implied by the word "improvements"
You started this by objecting to my wording ("among the most") when I said fish/chicken are the most sustainable meat options. They are, by a wide margin. Beef’s footprint is roughly 10× higher, so swapping a beef meal for chicken or fish cuts ~90% of those emissions. That’s not a "slightly less bad choice".
Calling harm reduction "silly" because tofu exists just shifts the target. We can hold two thoughts at once: (1) plant-heavy diets are best, and (2) for the vast majority who aren’t going vegan tomorrow, steering from beef to chicken/fish dramatically reduces damage right now. Dismissing that because it’s not maximal purity guarantees we leave real cuts on the table.
I agree it’s worth comparing beef sources! That was my point about within-category differences and harm reduction. Saying "tofu is cleaner" doesn’t make beef comparisons pointless - just like the existence of bicycles doesn’t make car fuel economy comparisons pointless. We should compare across categories and within them, so people who aren’t switching today still choose the lower-impact option.
I don't think it's "unserious" to recognize that >85% of the world's population eats meat.
If you're quibbling about wording, all I meant was: farmed fish and chicken are among the most sustainable meat sources.
I'm not making a statement that people should eat meat, but many people do eat meat, so it's worth comparing which meat sources are better than others. I think it would be great if more people knew that beef produces 10x the greenhouse gases than chicken/fish do.