I think the defining moment is that IF China gets man back to the moon before the US, that will be the moment when it becomes too hard to ignore. That is the new political high ground.
The Chang’e missions plan for this in 2030 and they have hit every target to date. While the US is currently targeting 2028, it isn't inconceivable that it can slip by two years.
Anecdotally, lower margin industries (think manufacturing sector) recruit almost exclusively immigrant tech workers because the salaries are too low to retain any but the least skilled US coders. The naive interpretation is that these companies would pay higher wages without access to H1B workers. Studies like TFA suggest that is not actually true. Companies in all sectors are constrained by the availability of skilled labor and to some extent are paying what they can afford. Reducing the number of H1B visas would likely hurt employers to the extent that their workers would suffer. It is not sensible to pretend that all of those jobs would still be there if you cut the H1B program.
Well, not quite the classic calling card - LLMs generally use a contrast of opposites. (e.g. "it wasn't a bug -- it was a feature", although that predates LLMs :)
TFA's use is more common in "normal" language: "it's not just [minor], it's [major]". (But, as others have pointed out, it was probably deliberately parodic anyway.)
Alas, at present, AI mostly serves to fill my email folder with vapid praise that clumsily restates the back-of-the-book blurb and then tries to get me to give someone else money for dubious (nonexistent) marketing services.
True. There is at least one more case: when LLM the other person is using has access to their context and information repositories that they don't want to share directly, and so the LLM text gives a peek at a slice of that, and that's not something I can recreate from thin air. In that case the LLM text may have utility for me.
There was (and still is) a bunch of niche Linux kernels. I remember reading about a team that used Xen to do triple-redundancy with voting for Linux. They virtualized _all_ the non-deterministic IO paths (including rdtsc) and observed the outgoing network packets, ensuring that they are completely identical. I think this was around the mid-2000-s timeframe.
But yeah, it was all super-niche. The consensus now is that you should design your systems to be fault-tolerant and self-recovering, rather than depending on perfect hardware functionality. And for everything else IBM still exists.
i ran some interesting experiments on jev today and wanted to share the results. i compared cost and latency of support ticket triage with different arrival rates for jev vs mainstream LLMs: https://suraj-website-eta.vercel.app/blog/what-a-correct-dec...
I don’t know what you’re saying, but for those confused. Starting in 2025, the de minimis exception was removed. This added delays to purchases of drugs from Canada to the US, and required consumers to pay new duties. These duties were collected by the pharmacies at order time, or at least my experience was that the pharmacies would charge your card for them. This is the common sense understanding of what I said above.
These duties were annoying but bearable. However these new changes will effectively shut down all consumer orders of drugs from Canada starting on Oct 22. So if you need these drugs to stay alive, order them right now.
I wrote a small proxy that points points to a jev api and a frontier api.
My harness connects to the proxy and only sees the frontier api, models, commands, etc. When I send prompts with tool calls, proxy routes to jev, jev narrows the tools, proxy cleans/sends to the frontier api.
So far in my tests, about 60% less tool calls. I'm also going to implement model switching, so it can use cheaper models. I think my workbench harness needs its prompts cleaned up.
> It's not like mathematicians are doing mathematics just for the funsies.
But they do. Most of higher math has no practical applications and is just a mental playground, philosophy constrained by formal logic and a set of axioms.
Hope is these capabilities will somehow translate to something more practical like physics, chemistry or biology.
Hey. Thanks for reading. From what I understand, the Mazda MX-5, which I believe you're referring to, is more of a two-seater sports car than a subcompact or compact passenger car, which are the segments the article focuses on. Also, the MX-5 is still in production as of 2026: https://www.mazdausa.com/vehicles/mx-5-miata.
A better example of a Mazda model I failed to mention would be the Mazda2, which was a subcompact that discontinued in 2014 (https://www.cars.com/research/mazda-mazda2/). Thanks for the nudge, though. I just realized the Mazda3 (a compact passenger car) is still in production, so it should be on the list of small cars that are still around.
> AI has different abilities, but it also has different weaknesses. It's straightforward to accurately log all of its behaviors, and you can even re-run it to see what it would do in myriad situations.
In principle that sounds true, but your specific example was already disproven in practice. Look at the Hugging Face hack - the amount of logs collected is so massive, that no one is even approaching this without using LLMs to help sift through them. With current models we're already way past being able to keep up with the volume of behavioral logs, and that's for post-hoc analysis; for real time defense, we already rely on classifiers (read: weaker LLMs and different ML models) to do this job.
Here’s a reminder that every time you use Grok, you enable the rise of a megalomaniac white supremacist who has killed a million people across Asia and Africa. If you don’t like hearing that, maybe ask yourself where your moral compass went.
Once AI is good enough the AI will control what I see, plus, I will just spend a lot less time looking at screens altogether.
It's also trivial for Amazon competitors to pop up in this potential AI future. Maybe you only sell door knobs, but all you need to do is set up a mostly text-based API designed for AI consumption and you're in business. If you offer the door knob for 10 cents less, why wouldn't the AI agents buy from you?
Right now making an Amazon competitor will fail mainly because humans are lazy and can't be bother to check other sites. AI might change that.
Finetune and infer: One instance ModernBERT didn't have the learning capacity for a single problem in the shape of my subjective preference task with finetuning, do you not have the basic research taste to realize no conceivable post-training recipe will result in an instance that can zero-shot hundred plus similar questions that vary with each sample?!
You really need to try that to find out?
And again have you actually tried Jev? It has a ton of world knowledge: it's able to infer user personas based on TV show watch histories using fairly recent titles... where the hell do you think that capability is emerging in 395M params?
The irony is if you really want to die on this hill, there are much better angles by focusing on LLMs that've had diffusion heads attached for fast inference with as much of a constrained decoding intelligence penalty: at least that'd put you in the ballpark.
I was being charitable that you know the field and are clueless about Jev, mea culpa for giving you the space to think I'm the one that's missing something.
The Chang’e missions plan for this in 2030 and they have hit every target to date. While the US is currently targeting 2028, it isn't inconceivable that it can slip by two years.