There's a lot of inference providers. This is an attempt to map each provider to the state's it's most like, not (necessarily) where it’s based. AKA, it's a vibes map for inference.
The best open models today have been “good enough” to do most knowledge work for a few months. This might be the first US one good enough too. That’s a big deal.
Things have been remarkably "not terrible" since approximately the release of Laguna S 2.1 a while back in early summer 2026, which is now outdated within the past 6 month time frame by other, better non CN open weight models.
I'm not saying that any of them are to the level as a deepseek v4.1 flash 1017 or similar but they're catching up. The most recent huge mistral release is also grounds for an optimistic view.
It's not as good as the current frontier, to be fair. I don't use it to help me think through architecture, for example. But I have found to work well for at least like half of coding tasks once architecture and patterns are in place. Its advantage is that it's very very fast.
MoA schemes that use proposers/task-separation/supervisory roles can still benefit from very fast diffusion models like mercury, the overall speed doesn't have to be bottle-necked by the slowest model -- that depends on the arrangement it's deployed in.
but to your point , yeah diffusion models can be kind of brain dead without proper coaxing.
my ideal arrangement is a smarter/bigger model supervising lots of proposals from diffusion models. The diffusion models spit out a big proposal each, the bigger model can vet something as minor as the headline or filename for each proposal, time still ends up being saved.
Sorry to put it bluntly, but, Mercury is sh*t in comparison. And I have genuinely tried a few times to use it since the speed is so good. But nop.. it just isn't there.
“Junior” is the wrong distinction to make. I’ve recently worked with senior engineers who are terrible at wielding agents, and recent grad engineers who multiple themselves 100x. The main way I look for software engineers now are:
- Deep curiosity and ability to absorb new information quickly
- Evidence of ability to wield and coordinate agents to the max
- Ability to clearly communicate ideas
You have these skills, you can conquer the software world.
If LLM's are supposed to get smarter, wouldn't they be able to wield agents for you? Seems like a skill with a short shelf life. In my day to day work i just use claude and it seems to spin up agents as it sees fit so I'm not sure why people need to coordinate agents to the max? If there is one thing i've seen in software engineering over many years is that sometimes software engineers prefer using flashy stuff even though it doesn't have that much benefits but they evangelize this stuff like it does. "Oh you don't use micro services and X,Y,Z for your crud app?", "Oh you use VS code and not vi?" "Why don't you maintain a data platform for your small scale data that can fit in postgres" so on and so forth.
Valid overall, but text editors like vi (or emacs or whatever else you prefer in the category) really do have major benefits and are definitely not being chosen to be "flashy". Skilled users choose one that resonates with them, and practice using it specifically. Traditional mouse-oriented editors just don't deliver that UX. It's not at all like choosing a bloated or over-engineered project architecture.
I've used vi and emacs and the number of their users that really 10x everyone else is vanishingly small. A large percentage of their userbases are people who wished they were the 10x but were much slower (or at best equal to) than the average Jetbrains tool users.
> If there is one thing i've seen in software engineering over many years is that sometimes software engineers prefer using flashy stuff even though it doesn't have that much benefits
We used to call this RDD - Resume Driven Development.
I’ve hired quite a bit over the last 10 years. Curiosity is the absolute number one thing I look for, whether for a fresh grad or a principal/staff+
The difference between a junior and a senior is still very real. With agents, knowing what to ask them to do (or not do) is bordering on more important than it ever was. A curious junior might still go down rabbit holes for things that a person who had done this before. That’s what makes them a junior. Not everything we _can_ do we _should_ do
I’d hate to be on a team with 100x junior output, yikes. Do you think seniority comes from being old or from being mentored into best practices and flexibility of pattern-matching?
Where I work we make real hardware and software that are used by real people doing real jobs every day. We use AI but in that environment someone who made Jackobriens comment would be seen as an idiot, someone who doesn't have a clue how to build software with real users that has to work with few bugs and be maintainable.
But there are so many people who make statements like his that maybe we should assume that some of them are not idiots or bots. If he's not an idiot it means that he's working in an environment where that kind of statement is paying off. So it could be, for example, that he doesn't really have profitable customers. His main 'customers' are investors who are clueless, or who know that they are clueless but don't care because they are going to move it on to a greater fool anyway. Or that because there aren't any customers, the only thing they can measure is 'productivity' in terms of 'number of unused, unprofitable and unsustainable features'
I work with AI as well and it can help, but it can also hurt. I think any objective person agrees with that.
There is another option to consider. What if they heard someone else say it and they believed it.
Now they go around, like many insecure people do, shouting about their unfounded take. Begging the universe to correct them or give them a good reason why they arent informed.
Then what usually happens with people who suffer from this type of short ordered thinking is, they realize they can't walk it back without social consequences because of the shouting they have done. So they just double down on it and find ways to live with the cognitive dissonance. Anthropic, openai, etc make billions off of people like this.
Because they propagate the BS advertising to people who don't know better, and then spend more money to try to make it real.
Well anthropic says you are wrong and wants juniors to disappear into another trade. This absolutely wouldn't be a coordinated effort to deskill Americans and force companies to helplessly accept whatever LLM code is generated without human review. Most certainly not. They wouldn't target regulated industries that require certain work praxis like biotechnology and lobby heavily with the military and government to change laws around this. Definitely not!
What would Anthropic gain by wanting juniors to disappear into another trade? Obviously their goal is to turn a profit meaning they need to be more efficient and they seem to believe that will come from senior engineers. They have the supply for it, why hire juniors when seniors are dying to working in your shop
Just because Anthropic isn't hiring juniors doesn't mean the rest of the world wouldn't or shouldn't. They have pressure to be the best on an extremely competitive surprisingly easy to enter market, they don't have time for juniors.
This reminds me a bit of the gaming industry. In most AAA game studios, junior openings are very rare or simply non existent. Companies expect you to learn the ropes elsewhere and only show up when you're experienced.
As long as there other companies scooping up the juniors (which we cannot assume will happen), this could be sustainable.
> Companies expect you to learn the ropes elsewhere and only show up when you're experienced.
This is every industry - barring unionized trades with established apprenticeship pathways - over the past few decades. Either someone does the early-career job-hop dance, or they take on student loans to (maybe!) skip some of that. (And, you know, universities would prefer not actually to train anyone, either. If they can - "highly selectively" - find cohorts who don't need much mentorship then that's much better for their bottom lines.) Regardless, the entire cost and risk of training has been pushed onto individual employees.
Yes, this works well for a small minority of highly-motivated people - who are, of course, vastly over-represented on this board, so few here will object to it - but it prevents far, far more people from ever reaching their potential.
Uh complete market capture for an entire industry that costs billions of dollars a year. They have been claiming software engineer was solved for years now.
If everyone does it, no one monitors it anymore, and quality degrades the only way out is more tokens.
Bingo. From day one what these companies have really been after is not "intelligence" but rather mass production and the concomitant quality expectation and behavioral shifts that go with it.
All they really need to do is convince enough people to adopt lower expectations and accept worse software products. At that point, mission accomplished. They've annihilated the source of value originally provided by people who actually know how to do things.
People here do speak as if majority of software is quality. Most software is meh at best, I am willing to bet a fully autonomous opus 5.5 would create and maintain software better than the average software companies build and force people to use in house.
These guys are awesome and I’m really impressed. The model alone looks great (according to benchmarks), and very cool they’ve done so much in building out the harness of top of it all.
They already use fiber-optic borescopes with no electronics on the lens side to create distance from electronics when their targets do sweeps.
You can also have opto-mechanical listening devices where the electronics are 100 m away. There is a vibrating membrane at one end, and a laser through the fiber reads the vibrations.
I like the story, I really can’t stand the way you used AI to write this. The story is great! But I genuinely cannot tell what you actually think. Too much over-wording and cheesy phrasing.
This was nowhere near the top submission. But even if a solo engineer could get a top kernel, you don't think that having thousands of engineers, infinite tokens, and stronger models than are available to the public would give the labs a significant edge?
Does the edge matter? I know you added significant as your hedge, but once you have feedback, your gain is largely irrelevant. Gain buys you bandwidth, so we are constructing systems run by the most powerful corporations where they are now optimizing for latency, as Archer says, do you want to flash crash civilization? This is how you do it.
1. labs have lots of inference capacity
2. they will have domain experts working on this so their efficiency is gonna be exponentially more (can direct LLM better, save money, reach same results faster)
You can't exceed roofline performance on hardware. There is an performance cap you can hit. This recursive self improvement stuff lets you be closer to the pareto frontier, but the idea that it is leading to some exponential growth is a total pipe dream.
Yeah, but here's a dirty little secret that very few people are discussing:
You can't use Claude for this sort of thing if the goal is to make better AI systems. Anthropic finetunes Claude to dissuade people and the agent from using research that actually works. Anything that they use internally in their own models is poisoned, to protect their moat.
By proxy, that also means any openweights model that was distilled from Claude is equally useless for this purpose.
Thankfully, I don't believe OpenAI does this - they are far more honest and seem to care about their reputation. Anthropic is evil though.
OpenAI is run by its employees and their culture is far healthier. Unlike Anthropic, they don't have a CEO that actively encourages their employees to be dishonest.
reply