Once a spec becomes sufficiently large and detailed and complicated, it becomes very difficult to ensure it is internally consistent. That's why I start every project with a METASPEC.md so that Claude can break up the task of writing SPEC.md into manageable steps.
They aren't saying that other models have the same overall level of capability. They are saying that the specific capability that the US Government tested is also available in other models.
20 years ago everyone was sold css entirely on the premise that, once the standards were adopted by all of the browsers, we would all be writing purely semantic html with completely orthogonal and swappable css. And today literally no one designs web sites that way - html today is mostly specific to presentation. It feels like pretty dramatic technological failure to me.
For Vim it isn’t replacing mouse necessarily. It’s giving you another way to navigate the cursor around the buffer by giving you absolute references rather than relative motions.
“Fed up” as a phrase comes from feeding livestock up to their fill. It’s very similar to how you would say “filled up”. So the upness comes from raising the level up to the limit.
For a long time I thought that the AdSense business model was ultimately doomed because I assumed that people hate ads as much as I do. It turns out I was just wrong about what most people are willing to put up with.
While I'm really enjoying this paper, I think you are way overstating the significance here. This is mathematically interesting, and conceptually elegant, but there is nothing in this paper that suggests a competitive regression or optimisation approach.
I might have misunderstood, but from the two "Why do X when you can do just Y with EML" sentences, I think you are describing symbolic regression, which has been around for quite some time and is a serious grown-up technique these days. But even the best symbolic regression tools do not typically "replace" other regression approaches.
I don't quite understand what is going on with the "spotlight" UI concept - I can click around on the characters and it highlights an area and it also reloads the landscape local to the character that I clicked on, so I can sort of traverse the similarity landscape this way. But I feel like I might be missing some part of the visual metaphor?
I think your assumption about inlining is essentially correct. As far as I know postgres was the last major rdbms to have an optimiser fence around CTEs.
I've been working on a side project for ~10 years (very intermittently) that involves a tricky combination of mathematics, classical AI algorithms, and programming language design, and I've gone though this very slow but rewarding journey to work out how all of the pieces should fit together properly.
In the last year or so I've been able to prototype it and accelerate the development quite significantly using Claude and pals, and now it is very close to a finished product. One one hand there's no doubt in my mind that the LLM tools can make this sort of thing faster and let you churn through ideas until you find the right ones, but on the other hand, if I hadn't had that slow burn of mostly just thinking about it conceptually for 10 years, I would have ended up vibe coding a much worse product.
Openclaw has 20k commits, almost 700k lines of code, and it is only four months old. I feel confident that that sort of code base would have a no coherent architecture at all, and also that no human has a good mental model of how the various subsystems interact.
I’m sure we’ll all learn a lot from these early days of agentic coding.
I think the total volume idea is more flawed than you realise. Pretty much everyone would be able to achieve greater volume, on any exercise, just by decreasing the weight, so your high rep caveat is covering up for quite a lot. This is true mathematically for an Epley style model for example.
To add to this, I find talking to it about code quality or architecture issues can work quite well. Just treating it like another developer. Saying, “I’m not happy with the way the project is going because of X, and Y” and then making a plan for how to get things back on track. Maybe putting a complete rewrite on the table, or maybe just having it record the agreed code style principles in CLAUDE.md, etc
I think I have almost the opposite intuition. The fact that attention models are capable of making sophisticated logical constructions within a recursive grammar, even for a simple DSL like SQL, is kind of surprising. I think it’s likely that this property does depend on training on a very large and more general corpus, and hence demands the full parameter space that we need for conversational writing.
Concrete Mathematics is probably the best single book that you could read to prepare you for some the problems beyond the first 50. It’s extremely fun, and also mathematically serious. A large portion of PE problems are exactly in the cross sections of number theory, combinatorics, and computation that is covered in this book.