_jab

Both can be true simultaneously. Anthropic can probably be trusted not to train on our Fable sessions, but eroding ZDR as the industry standard still sets a dangerous precedent.

There's a parallel between data retention and general mass surveillance. Sure, both systems can be used for purely benign purposes, with appropriate safeguards in place. But history shows that surveillance systems are alarmingly easy to co-opt for nefarious means, and model providers do have a heck of an incentive to leverage retained data for internal means.

This is worth protesting, even if I believe this policy itself does not immediately compromise my privacy.

It's pretty simple; organizations are willing to tolerate paying $1500/month/engineer, which seems to be roughly inline with "normal" consumption for most full-time engineers. If that number grows significantly, then I bet companies will start exploring flash models more, as you propose.

While $835k is undoubtedly a lot of money for this man, split among Tennessee's 7M residents, this works out to be less per taxpayer than the sales tax on a latte.

Still, this idea bears merit for other reasons. Americans routinely underestimate how much money is spent on Social Security, healthcare, and debt payments, and overestimate how much money is spent on education and infrastructure. More clarity into that could help build real political momentum to actually balance the budget.

This agreement feels so friendly towards OpenAI that it's not obvious to me why Microsoft accepted this. I guess Microsoft just realized that the previous agreement was kneecapping OpenAI so much that the investment was at risk, especially with serious competition now coming from Anthropic?

Vercel did not specify which of its systems were compromised

I’m no security engineer, but this is flatly unacceptable, right? This feels like Vercel is covering its own ass in favor of helping its customers understand the impact of this incident.

With intense competition for enterprise contracts coming from Anthropic, I thought this was OpenAI's time to get _less_ memey, not more. What the hell are they thinking?

Between the rounded corners that don't reach the edges of the viewport, and the behavior when opening a new app for the first time, it feels like Mac's UI is optimized around the assumption most users won't expand windows to fill the whole screen, but rather leave them half-sized somewhere in the middle.

Does anyone actually do this? Especially for heavy-duty applications like my web browser and IDE, this has always felt like a bizarre assumption to me.

Sure, but it's not as if the DoD was planning on using Anthropic to _collect_ the data either? I assume that the hypothetical DoD use case Anthropic shied away from dealt with the processing of surveillance data, just like what Palantir does.

I've found current-generation Macs so capable that I've switched to using a Macbook Air. Would strongly recommend - it's still a powerful machine and it's significantly lighter and cheaper.

Completely agree.

Crazier question: what’s wrong with a well-intentioned surveillance state? Preventing crime is a noble goal, and sometimes I just don’t think some vague notion of privacy is more important than that.

I sometimes feel that the tech community would find the above opinion far more outlandish than the general population would.

I'm pretty tempted to discredit this article on the basis of the author's lack of legal expertise, but to be honest I don't really have the expertise to properly comment here either.

But I don't think the author is correctly interpreting the principles of legal ethics, and their repeated questioning of attorney-client privilege, which I've considered to be one of the foundations of the American legal system, is hard to take seriously.

Also, I don't think their depiction of John Adams's representation of the British soldiers is accurate. From what I can tell, Adams sought only to give his clients as strong a legal defense as possible. In the trial, he called the American protestors a "mob", gave a racist depiction of one of the victims to justify the soldiers' panic, and ultimately saw all but two soldiers acquitted. Adams viewed this as a patriotic act, yes, but only insofar as he believed all accused of crimes in America deserved fair legal representation. He was a lawyer defending his clients, not the judge or jury trying to find the "truth" of the matter.

I've often wondered whether the world would be better without ads. The incentive to create services (especially in social media) that strive to addict their users feels toxic to society. Often, it feels uncertain whether these services are providing actual value, and I suspect that whether a user would pay for a service in lieu of watching ads is incidentally a good barometer for whether real value is present.

Don't get me wrong, I'm well aware this is impractical. But it's fun to think about sometimes.

Next.js is still pretty damn widely used.

This argument doesn’t make much sense to me. Claude Code, like any product, presumably has dozens of external dependencies. What’s so special about Bun specifically that motivated an acquisition?

The logical conclusion here would be to have no door for the bathroom, but to have specifically the toilet in a separate subroom.

But I don’t think this makes much sense anyways. The hotel industry is not one that thrives from repeat patronage, and “the bathroom has no doors” features rarely in marketing.

Programmatic tool invocation is a great idea, but it also increasingly raises the question of what the point of well-defined tools even is now.

Most MCP servers are just wrappers around existing, well-known APIs. If agents are now given an environment for arbitrary code execution, why not just let them call those APIs directly?

GitHub is pretty easily the most unreliable service I've used in the past five years. Is GitLab better in this regard? At this point my trust in GitHub is essentially zero - they don't deserve my money any longer.

One symptom of AGI fantasy that I particularly hate is the dismissal of applied AI companies as "wrappers" - as if they're not offering any real technical add on top of the models themselves.

This seems to be a problem specific to AI. No one casts startups that build off of blockchains as thin, nor the many companies that were enabled by cloud computing and mobile computing as recklessly endangered by competition from the maintainers of those platforms.

The reality is that applying AI to real challenges is an important and distinct problem space from just building AI models in the first place. And in my view, AI is in dire need of more investment in this space - a recent MIT study found that 95% of AI pilots at major organizations are ending in failure.

In the US, one of the highest farebox recovery ratio transit systems has historically been BART, which is 2019 was 72%, and even today is around 50%.

Unfortunately, having a very high ratio also makes systems much more vulnerable to collapse during periods of economic downturn, which is exactly what BART has been dealing with since ridership collapsed during Covid.

I'm no expert in this topic (in other words, I just asked Claude this), but AFAICT part of the reason Japanese rail systems did better appears to be that they are owned by diversified companies that own numerous other assets, like hotels, restaurants, and office complexes.

Many questioning why Microsoft would agree to this, but to me the concessions they made strike me as minor.

OpenAI remains Microsoft’s frontier model partner and Microsoft continues to have exclusive IP rights and Azure API exclusivity

This should be the headline - Microsoft maintains its financial and intellectual stranglehold on OpenAI.

And meanwhile, while vaguer, a few of the bullet points are potentially very favorable to Microsoft:

Microsoft can now independently pursue AGI alone or in partnership with third parties.

The revenue share agreement remains until the expert panel verifies AGI, though payments will be made over a longer period of time.

Hard to say what a "longer period of time" means, but I presume it is substantial enough to make this a major concession from OpenAI.

Gotta be honest, I think the spoon bending metaphor is unhelpful, and only misleads the audience and buries the lede here. It took me a while to figure out what this repo actually does.

But the insights are indeed interesting. I'm curious if you've found any way to quantify alignment differences between GPT-5 and the previous generation?

I'm skeptical of arguments like this. If we look at most impactful technologies since the year 2000, AI is not even in my top 3. Social networking, mobile computing, and cloud computing have all done more to alter society and daily life than has AI.

And yes, I recognize that AI has already created profound change, in that every software engineer now depends heavily on copilots, in that education faces a major integrity challenge, and in that search has been completely changed. I just don't think those changes are on the same level as the normalization of cutting-edge computers in everyone's pockets, as our personal relationships becoming increasingly online, nor as the enablement for startups to scale without having to maintain physical compute infrastructure.

To me, the treating of AI as "different" is still unsubstantiated. Could we get there? Absolutely. We just haven't yet. But some people start to talk about it almost in a way that's reminiscent of Pascal's Wager, as if the slight chance of a godly reward from producing AI means it is rational to devote our all to it. But I'm still holding my breath.

AAVE is definitely underappreciated as the source of a lot of common modern slang. But in this case, the article makes it pretty clear that "try and" is not nearly modern enough to have come from AAVE - they show several attestations from the 1500s and even mention one from 1390.

I've found that there's a pretty consistent relationship between how clearly I can imagine what the code should look like, and how effective vibe coding is. Part of the reason for that is that it means I'll be more opinionated about the output of the model, and can more quickly tell whether it's done something reasonable.

I suspect practicality is not the goal here, but rather a proof of concept. Perhaps they saw speed as an important technical barrier to cross

Sometimes when I give simpler technical questions (applies to system design too), candidates begin to massively overthink it. The question seems so simple that they start anticipating high-scope extensions that don't actually exist, like turning a basic algorithm into a distributed, high-availability pipeline. There's only so much you can do as an interviewer to rein candidates back in at that point, and those interviews tend to go off the rails pretty quickly, with little code actually ending up being written.

Thing is, I reckon those candidates aren't a good fit anyways. The biggest mistake I see engineers who join startups make is that they pursue excessively sophisticated solutions, with robustness that does not justify the added complexity. I'm sure these candidates are smart, but they're not good fits for the high-velocity, simplicity-obsessed technical environment of a product-play startup.

Anthropic is saying that one out of every 20 users will hit the new limit.

Very good point, I find it unlikely that 1/20 users is account sharing or running 24/7 agentic workflows.

There’s a lot of ideation for coding HUDs in the comments, but ironically I think the core feature of most coding copilots is already best described as a HUD: tab completion.

And interestingly, that is indeed the feature I find most compelling from Cursor. I particularly love when I’m doing a small refactor, like changing a naming convention for a few variables, and after I make the first edit manually Cursor will jump in with tab suggestions for the rest.

To me, that fully encapsulates the definition of a HUD. It’s a delightful experience, and it’s also why I think anyone who pushes the exclusively-copilot oriented Claude Code as a superior replacement is just wrong.

The company should have been worth at least the cash it had on hand, which has been reported as ~$100M. It's also been reported that all vested equity and VC shares were bought out (although apparently perhaps with a few exceptions for people who declined the offer), which meant that the employee unvested equity stakes were "undiluted" from whatever they were before (hard to judge, but maybe 5-10%), to 100%. So every employee had their stake in the company increase 10x-20x. So if the company had then decided to simply close up and distribute the remaining cash as dividends to the employees, it would be as if each employee had simply been bought out pre-deal at a $1-2B valuation. And that was the absolute worst case scenario - clearly Windsurf found a better deal with Cognition.

The details here remain unclear to me, and even this tweet is somewhat vague.

I was given an offer that would explode same day. I had to forfeit all of my vested shares earned over my 3.5+ years at Windsurf. I was ultimately given a payout of only 1% of what my shares would have been worth at the time of the deal.

Was forfeiting the vested shares conditional on accepting the offer, or did he have no choice over the matter? Was the payout what he was offered as part of accepting the deal, or was that his consolation for not accepting it? The wording is genuinely unclear to me.

I literally see 3 interpretations here:

1. Offer was to forfeit shares in exchange for 1% payout, but OP rejected and still has shares

2. Offer was to forfeit shares in exchange for undisclosed payout, but OP rejected and got 1% payout instead and still has shares

3. He had to forfeit shares regardless of accepting offer, got 1% payout

(1) and (3) are both shitty offers from Google, but (2) is reasonable. Exploding offers are not uncommon in tech acquisitions. My guess is that (2) is what happened, since that's not in contradiction with prior reporting.