resters

Perhaps, but things like Sora burned a lot of compute -- a high percentage.

While I completely agree with your take, I think everyone has been surprised by how quickly LLMs have become highly useful and extremely powerful, and by how possible it is for relatively smaller models to also be highly useful.

Given that, I would expect that in hindsight OpenAI and Anthropic would spend 40% of what they have on compute if starting over and knowing the actual landscape.

The massive capital allocation was a blind decision and they swung big.

It is still possible that techniques will be developed that create a moat where the massive hardware capex is justified, but US policies of banning competitive GPUs and blocking frontier lab releases makes such things far less likely.

"Escape velocity" for AI is when the open weight models are good enough to help drive the next frontier innovations/techniques. I think we are close to that if not already there, at which point it's a race to commoditization no matter what Altman or Lutnik wish will happen.

How is merging identities the solution?

The goal is not "how can I do blackmail-worthy activities without being blackmailed?" the goal is "how can I participate in a narrow way without, for example, having those reading my content biased by my LinkedIn profile or my corporate affiliation?"

We don't walk around wearing little badges that identify every organization we affiliate with, every group, etc. Why would anyone want their online participation burdened by all that extra (usually irrelevant) information?

In my case, I don't write content I would worry about some determined prose fingerprinter finding, and if I did I would sanitize the prose first :)

Does it really matter that a review was written by "Ronald B Smith" vs "BrewersFan4000"?

I'd argue that it only matters when the goal is to convert "real name" accounts into something that is effectively merged with an Experian soft credit score, Google supercookie-powered profile, etc.

There are some of us (late generation-X / early millennial) who saw this coming and still maintain a variety of separate identities across many domains.

I don't know why someone would want to have the same identity in the workplace as on internet forums, for example.

Social media appears to have given many people the idea that they ought to cultivate their public identity from an early age as preparation for internet fame / personal branding.

I'm very pleased to see open source mapping/navigation systems. I have had the hypothesis for a while that many of the UI/UX designers on the google maps team do not actually drive a car.

I make code review and PR approval optional for my teams. A solid process should emerge from respect and peer pressure alone.

This article is sort of a self-advertised red flag that the writer is rationality-challenged.

1) many disagreements are not ultimately about facts but about intentionally different tradeoffs/prioritization.

2) if in fact one argues on facts/logic then losing the argument means you had your own logic or facts corrected, which should be a good thing, not a bad one.

They are focused on the things you do when you are not over-capitalized and you can’t get unlimited nvidia hardware to train on. And the results speak for themselves.

Meanwhile we in the US are blocked from buying Huawei GPUs and retirees are boasting about the nvidia in their portfolios.

Thank you so much to everyone at DeepSeek who is working on this and who have the courage and generosity to open source this for humanity.

We in the United States will never forget!

For all the harm Trump does to the US at least he is helping China!

China’s leadership is pragmatic, transparent, mature.

Regulation at the state level (which is what Sacks and the toadies were so against) is far superior to this bc it is local and bad ideas only affect those in one state!

Bad Federal regulation has a much larger blast radius.

Let’s hope Chinese models save us from this. BYD is trying to save us from Elon but 100% tariffs are welfare keeping Elon afloat.

This is what we get when the president is a nepo baby who inherited a fortune and bankrupted many companies and had mainly business failures before entering politics.

I’m quite confident that few people are more ignorant on AI than Donald Trump. That he promised less regulation and lied is no different than his many other lies.

In six months everyone will cancel frontier subscriptions.

You're right it's not only trade policy, but I think most of the fab contracts on current models were already negotiated and Apple ate $3.3B of tariffs as a COGS increase (delaying passthrough avoids spotlighting tariff-driven inflation). Increasing DRAM prices are a factor, but would not be a 20% BOM price increase at all (much less on the total price) for most of the impacted devices. The magnitude and the simultaneous across-the-line timing look more like margin recovery than a component passthrough.

Expect this trend to continue -- firms have delayed price adjustments to avoid retaliation from Trump as doing so would draw attention to Trump's many inflationary policies.

Now all of the businesses who use Apple products as an input are more likely to raise their own prices, etc. This is how inflation happens across the economy. Trade war leads to price increases on Apple's inputs, Apple has to raise prices, etc.

to train models to be smarter than they are, one needs examples and cases to train on, and once you get close to the top percentiles of human reasoning there is extremely little such material available.

You can create contrived logic problems, but they often turn into language games because English is not formal logic.

And you can train on "monty hall" style problems, but those too are language games that are intriguing to humans but obvious when framed slightly differently.

In other words, model trainers are fighting against the overwhelming mediocrity of the training corpus (all of the recorded human output from history).

As models improve, the next phase will be models co-designed with humans to overcome these limits. The way we use language and the process we use to problem solve (we currently call this "orchestration") will evolve as part of this. Meatspace metaphors map badly when we have massive context and don't need the same limits. How different is hallucination from extrapolation, etc.

Much of the skepticism and confusion about LLMs is no different than a person of average intelligence hearing a highly intelligent person explain something and considering the explanation gibberish, then arrogantly accusing the intelligent person of being unhelpful.

Much like dogs were domesticated from wolves to have traits that make them good around humans, LLMs will evolve around our limits, around our arrogance, around our aesthetic biases and prejudices. Intelligence and rationality is fundamentally not what most humans want from an LLM.

It's not surprising if one has the mental model of the probability that the request gets enqueued. Then when you add variable time to process requests it becomes more clear why some requests can take unexpectedly long (there is a >0 probability that a request gets queued behind several of the slowest endpoints, for example). So even if 90% of the endpoints are fast and most of the requests aren't even queued, there will still be some that end up being quite slow.

It's not surprising that Trump is as bad as he is at many of these things, but what is surprising is that he's worshiped by his supporters.

Great point. I wish more people realized that intuitively.

I tried it and it was extremely broken and dramatically overhyped. Which I guess is what Elon finds appealing about it, as that seems to be his preferred type of business.

While there is some irony in the AI is dangerous marketing Anthropic uses, the main story here is that the Trump administration is apparently retaliating against Anthropic for refusing to relax certain safeguards. Trump and Hegseth have both posted highly immature, vindictive social media posts.

Most notably, any default assumption one might have had that the Trump administration can be counted upon to act in good faith should be viewed at this point as completely false. Even conservative legal scholars like Richard Epstein are shocked at the bad faith conduct across many areas.

This is a government making an authoritarian move to sabotage one of the top US AI companies. It's pure sabotage, nothing else.

Please point to a video or example of some way that Cursor is remotely state of the art at anything...

another competitive, American coding harness

Isn't the Cursor founder British?

Cursor was simply able to get early access to openai models and get an early lead doing things that are now obvious and done better by many others. Does anyone really want to use a crippled "enhanced" vscode to interact with a crippled version of codex or claude code?

I'm referring to the process described here: https://www.dhs.gov/cbphome

Sacks fraudulently obtained citizenship papers but the next administration will likely review the case and revoke his citizenship, which will make him eligible for deportation.

Sacks can choose to self-deport on his own, which will help his chances if he ever decides to re-apply for citizenship or entry into the US.

The government blocking access to frontier AI is the definition of AI dystopia. Let's hope Anthropic open sources Fable and teaches Donald Trump a lesson the way Deepseek did!

If Anthropic keeps Fable closed source and plays the Authoritarian game, at least we have hope that an upcoming Deepseek model will surpass Fable before long.

A few years ago Dropbox just stopped working for the basic thing it is supposed to do -- let me access lots of files without taking up tons of local drive space. Since then I have not stopped using/paying, but i have stopped counting on it and I have stopped adding new files to it, sharing files with it, etc.

The constant marketing for "dropbox for business" (which is priced badly and is not something I've ever felt comfortable recommending to any business) was also quite irritating.

Just as poorly designed code can still compile. This is operator error, not a failure of the technology.

This failure is human laziness, not an issue with the technology. People who use AI because they are trying to avoid doing work fall into a completely different category than people using AI as a force multiplier and for skills/capabilities enhancements / quality improvement.