alex_sf

My agent is currently doing the work of ~6 senior engineers, based on ticket closures. These are not trivial tasks: all of them involve a mix of code, judgement, executing remote commands in a production environment, etc.

This is at a well-known tech company operating at massive scale (and resulting complexity).

L1/L2 tech support is completely dead within a couple years. The delay is only around how long it takes people to realize.

1. that closed source models are more efficient than open source

Not a reasonable assumption for a variety of reasons.

2. Deepseek is served at a profit and not a loss

Not a reasonable assumption either.

Why do you need to know the architecture? Just compare Deepseek V4's performance with GPT 4 and treat internals as a blackbox.

Because the internals are what actually matter and what drives inference cost.

It would be entirely reasonable to expect that GPT-5.5 has some sort of optimizations or changes to the architecture to make it easier to train, or to make runtime ablation easier, or to better handle large batches, or whatever.

Those changes, particularly if they are non-public, can easily result in worse inference performance than a comparably sized model without those changes.

It is borderline conspiratorial to believe it this way.

It's not any sort of conspiracy. It's how land-grab tech companies have always worked. To presume otherwise is silly.

It's silicon valley and they are trying to aggressively grow. Your baseline assumption should be the exact opposite.

There's no reason to think that the latest frontier models have similar inference costs to open source models.

It would be more surprising if the surrounding architecture hasn't significantly diverged. If it _hasn't_ significantly diverged, then given the performance difference it would imply that the frontier models have significantly greater param counts, which would result in a higher cost.

The price a company charges, _particularly_ a high growth VC-backed one, is a poor signal for their costs.

That blog post is not very compelling either. Without knowing details of the architecture, comparing the various frontier models to open models doesn’t make sense.

I mean, they do (and did). If you aren't shipping, you'll be out eventually.

"shape" too, at least with gpt5.5, is coming up constantly.

It is in the same way that educated guessing is.

I guess (heh) it depends on your definition of 'educated guessing'? Looking at the problem, considering a solution, discarding it, trying another and testing, iteratively, is how most people would approach any tricky problem.

Brute force is substantially different. It would be saying that, other than maybe setting some basic bounds and heuristics, I'm going to try literally everything and test each. That's not at all what the LLM did here.

This isn't brute force.

Open models, in actual practice, don't match up to even one or two generation prior models from Anthropic/OpenAI/Google. They've clearly been trained on the benchmarks. Entirely possible it was by mistake, but it's definitely happening.

Which is why compilers decimated the software industry.

The same way companies already deal with any cost.

Most people just want something that looks nice. I understand it’s deeper to someone really into it, but the rest of us are fine with it.

Everytime I've tried a local model, and I have tried lots for a couple years now, they just seem like they were overtrained on benchmarks. They consistently perform dramatically worse than even older models from Anthropic/OAI/Google.

Europe is behind because we do not have good leadership. The decisions taken by leadership, no matter what level you look at - local, company, national, supranational - are rarely in the best interest of Europeans. Our markets - housing, rental, labor, capital, pension - are broken and therefore the population does not find opportunities to express their talent completely and the more motivated migrate. Europeans lack well-paying jobs and pay is low because pay is not transparent.

Sounds like Europe is behind because Europeans are working less and taking more vacations. You just point to poor leadership as the cause.

This isn't buried or hard to find, but in good faith:

https://www.dni.gov/files/ODNI/documents/assessments/ODNI-Un...

https://www.dni.gov/index.php/newsroom/congressional-testimo...

Saying "Accidents happen in war" is absolutely a way of saying "Accidents are acceptable in war".

Bridges fall down sometimes. I don't think it's acceptable. It's a statement of fact. There are always going to be mistakes, in every field and in pursuit of every goal. Your objection and implications aren't particularly charitable here.

My "brilliant" plan would have been the negotiations that were happening where Iran agreed to pretty strict monitoring and stipulations on nuclear fuel development.

Iran was not complying with the monitoring requirements.

The "Iran was getting nukes" rhetoric needs real evidence that was actually happening not "we think that might be happening because Trump said so."

Intelligence agencies under both Biden and Trump (and since at least the 90s) have repeatedly confirmed it.

This isn't really a question or doubt any reasonable person can have. There can be an argument about how close they are at any given moment, but they are actively pursuing nuclear weapons.

It's both.

If goods aren't being sold, then the price will drop.

That's not worth 45k. It's barely worth anything for a typical website, tbh.

Tbf, there is no one with a ‘serious DevSecOps background’. It’s an incredibly strong hint that the person is largely a goof.

Ralph loops are also stupid because they don't make use of kv cache properly.

This is a cost/resources thing. If it's more effective and the resources are available, it's completely fine.

If you are not too tired, drunk or using any substances, and not speeding, your chances of causing a serious traffic accident are miniscule.

You realize that like.. other people exist, right?

Would you sign up for such a system if you can volunteer to participate in it, with now those random killings being restricted to those who've signed up for it, including you?

I mean, we already have. You volunteer to participate in a system where ~40k people die in the US every year by engaging in travel on public roadways. If self-driving reduces that to 10k, that's a win. You're not really making any sense.

If the goal is to reduce the number of fatal mistakes, why is that argument garbage?

Taking RLHF into account: it's not actually generating the most plausible completion, it's generating one that's worse.

A fairly reliable determinant for how the Court will rule is found using a materialist analysis. That is, the Court will generally side with corporations and capital owners when given the choice.

This is a big claim. Do you have any evidence to support it?

In the wake of someone trying to prove the same for Congress, it was conclusively shown that the opposite was true:

https://www.vox.com/2016/5/9/11502464/gilens-page-oligarchy-...

I see several opinion pieces making the same claim, but no actual studies of their decisions.

More importantly: the concern can't and shouldn't be the income of the parties involved in a suit, but who is right and who isn't.

Rocky and CentOS are both based on Red Hat Enterprise Linux (RHEL).

CentOS used to be a free and open source downstream version of RHEL. Keeping the history short: Red Hat effectively acquired CentOS and discontinued it as a downstream version of RHEL. They turned it into 'CentOS Stream', which is, more or less, a continuously delivered upstream version of RHEL. This isn't acceptable for a large number of the CentOS user base.

One of the original founders of CentOS, Gregory Kurtzer, started Rocky as an alternative. It's basically what CentOS used to be: a free and open source downstream version of RHEL.

Lambda availability is awful.

It is definitely not an intended feature for the end user to be able to trick the model into believing it said something it didn't say. It also doesn't work with ChatGPT or Bing Chat, as far as I can tell. I was talking about the user, not about the developer.

Those aren't models, they are applications built on top of models.

That can be done with special tokens also. The difference is that the user can't enter those tokens themselves.

Sure. But there are no open models that do that, and no indication of whether the various closed models do it either.