nokun7

yes - kathack or something. I won a lot of friends because of it back in the day.

So much fun. If it had the original music that'd be super awesome!

What I mean is that later employees—especially the ones who joined during the 2021–2022 hype when Brex was valued at that crazy $12.3 billion peak—got their RSU grants priced at those very high levels. That meant their equity was basically "underwater" once valuations crashed post-2022; the shares they were promised wouldn’t pay out much (or anything meaningful) unless the company somehow got back to those crazy heights.

To keep people from jumping ship and to make things feel fairer, IIRC in 2024 Brex did some RSU "top-ups" - basically, they handed out extra shares at the much lower current valuation to compensate for the drop and give those folks a better shot at actually making some real money or "breaking even".

That's less than half of Brex's crazy $12.3 billion peak back in 2022.

But honestly, it’s still one of the biggest fintech deals ever and actually gives people real money in a market where most unicorns are just stuck. The founders are reportedly splitting about $1 billion each, early investors (2017-2018) are getting 12-80x returns, and YC’s tiny $120k seed turned into ~$100 million (800x, insane TBH). Even later folks (especially the 2021-2022 crowd) are breaking even (at least) or getting a little upside thanks to some 2024 RSU top-ups.

Yes! You bought back fond memories, despite the frustrations.

I used to play a game called Pro-Am on Nintendo. This game reminds me so much of that. The controls are basically the same.

What sorts of assumptions?

I user the ycombinator link, but it still says 0 credits, what gives?

> Building nôn pipelines

Never heard of "nôn pipelines" - is that like a new thing?

> She gets credit for being complicit in a violent overthrow of the Democratic government of the united States.

Care to elaborate what is this in reference to? Especially the "violent" part. Please do not simply link to some news article in your response.

While there is pingora https://blog.cloudflare.com/how-we-built-pingora-the-proxy-t..., I would recommend checking out envoy. Although not rust based, envoy has gained quite the repute for being extremely versatile and robust.

Basic question: Is PF related to ebpf by any chance?

I have seen Nextdoor also do the same and I had alway wondered. I believe it uses the same mechanism. Quite interesting.

I use ngrok for exactly this type of functionality. Can someone clarify why would anyone need malai over ngrok?

So, Finland gave fluoride a shot in a few cities back in the '60s through the '90s, but they didn’t see much point to it and dropped it. Now, some spots there are even filtering it out of the water. You can still get it in toothpaste and mouthwash, though.

There’s this study I read that says the top five countries for dental health are Denmark, Germany, Finland, Sweden, and the UK. The US? It’s lagging at ninth. Funny thing is, out of those top five, only the UK messes with fluoride in their water, and even then, it’s just for like 5% of folks.

Honestly, it kinda feels like skipping the fluoride in water might be the smarter move.

Honestly, while this is a great update and all, other AI platforms have had web search functionality for quite some time now. Any explanation for this delay?

I wonder if Claude’s API will match Perplexity’s dynamic answers. Is there API rate limiting. If so, then the older API pricing would be preferable. Can users switch between the two?

Aryabhāta, an exceptional Indian mathematician born in 476 CE, left a lasting mark on mathematics and astronomy. At the age of 23, in 499 CE, he calculated pi to be approximately 3.1416 and suggested its irrational nature, relying on insights from Vedic traditions. He’s also widely recognized for introducing zero as a numeral and developing various mathematical and astronomical concepts. That said, the value of pi had been explored even earlier by another Indian mathematician, Baudhayana, around the 6th century BCE. Baudhayana not only worked out pi but also laid out what we now call the Pythagorean Theorem, long before it reached European scholars.

Didn't parents care for their children before?

So then why do you think California - arguably the most liberal state - is also implementing a return to work mandate?

How timely!! I was looking for something similar to convert some of my portrait images. I'm trying it right now and I'll let you know how it goes soon. Thank you

What’s particularly intriguing is how these models handle uncertainty and potential "hallucinations". For instance, OpenAI’s o1/o3 have started hedging hallucinations more conspicuously, using phrases like “this likely contains ...,” which could serve as an alert for users to question the output. This evolution hints at an emerging self-awareness in AI design, where developers are training models to flag potential inaccuracies, potentially reducing trust issues in critical applications like coding or scientific research. However, TFA in the OP could reveal whether these safeguards can be bypassed or exploited, shedding light on the models’ vulnerability to adversarial inputs or creative prompting.

For Claude, inducing hallucinations might expose weaknesses in its "extended thinking" mode, where it deliberates longer but might overgeneralize or misinterpret certain unclear instructions. This could be especially relevant in real-world tasks like software development, where a hallucinated line of code could introduce subtle but costly bugs. Conversely, such experiments could also highlight the models’ creative potential.

Overall, an investigation into making these models hallucinate could push developers to refine safety mechanisms, improve transparency, and better align these systems with human intent, ensuring they remain reliable rather than unpredictable black boxes. This topic underscores the delicate balance between advancing AI capabilities and mitigating risks, a tension that will likely shape the future of AI development itself.

I can't disagree with that too much, although I have a very special place in my heart for the Intellij's of the world.

I recall when I used to work for Pupper Labs - then PuppetDB jumping to Clojure in 2012 was a gutsy call that totally paid off and put the language on the map. Puppet was a Ruby shop, but as data piled up, they needed something really cool and risky—enter Clojure, which we though was a slick Lisp running on the JVM. It turned PuppetDB into a beast for storing catalogs and reports, leaving Ruby’s old setup in the dust with crazy speed boosts, like 130 times faster for some tasks, thanks to its async magic and functional vibe. We could’ve gone Java or JRuby, but Clojure’s concise code and JVM power hit the sweet spot, making PuppetDB a lean, scalable backbone for thousands of nodes. It wasn’t just a tech switch—it showed Clojure could hang with the big dogs, influencing Puppet’s later projects and proving a niche language could rock real-world infrastructure. It was the best move and kept thing really interesting - in a good way of course.

So, Microsoft’s move to ditch leases for “a couple hundred megawatts” of data center capacity, as noted in TFA, is a pretty intriguing shift—and it’s not just a random cutback. Per some reports from Capacity Media and Analytics India Magazine, it looks like they’re pulling some of their international spending back to the U.S. and dialing down the global expansion frenzy. For context, that “couple hundred megawatts” could power roughly 150,000 homes, (typical U.S. energy stats) so it’s a decent chunk of capacity they’re letting go.

IMO it's not a full-on retreat—Microsoft’s still on track to drop $80 billion this fiscal year on AI infrastructure, as they’ve reaffirmed. But there’s a vibe of recalibration here. They might’ve overcooked their AI capacity plans, especially after being the top data center lessee in 2023 and early 2024. Meanwhile, OpenAI—Microsoft’s big AI partner—is reportedly eyeing other options, like Project Stargate with SoftBank, which could handle 75% of its compute needs by 2030 (per The Information report). That’s a potential shift in reliance that might’ve spooked Microsoft into rethinking its footprint.

Also it seems they're redirecting at least some costs - over half that $80 billion is staying stateside, per Microsoft’s own blog, which aligns with CEO Satya Nadella’s January earnings call push to keep meeting “exponentially more demand.” It’s a pragmatic flex—trim the fat, dodge an oversupply trap, and keep the core humming. Whether it’s genius or just good housekeeping, it shows even the giants can pivot when the AI race gets too hot.

As a longtime Vim user, I’ve got to say, the "completion-preview-mode" caught my eye. It’s pretty cool how Emacs is tossing in this slick, built-in predictive typing thing—kinda like Company or Corfu, but using its own minibuffer magic. Honestly, it’s a chill move that keeps things smooth without overcomplicating the vibe. Emacs just keeps doing its thing, polishing the edges while staying true to its hardcore, customizable soul—gotta respect that, even if I’m still Team Vim.

In my view, FlashMLA’s exclusive targeting of Hopper GPUs restricts its cross-platform use, and the lack of comprehensive documentation, vague compatibility with wider frameworks, and absence of benchmark comparisons or trade-off insights reduce its ease of use and adaptability. While it holds potential for specialists with tailored requirements, its specialized nature and limited community backing indicate it’s not yet a broadly practical tool, requiring more detailed guides and expanded hardware support to unlock its full capabilities.

Reminded me of a few similar articles I have read that reveal how divers’ real-world needs drive equipment evolution, transforming a basic human function into a catalyst for safer, more inclusive cold-water diving.

[1] "Why Do I Need to Pee Every Time I Dive?" (DIVER Magazine), and

[2] "Drysuit Diving Myths, Busted" (Scuba Diving)

> Imagine giving you car to strangers who will use it as a public trash can. You would need a hard societal change for people to respect other people's property.

Nothing that a bit of insurance cannot fix. But I do get your point.

The future of FSD in Tesla and other self-driving cars is looking pretty wild! Tesla’s been pushing hard with stuff like their Cybercab and unsupervised FSD plans, and if they pull it off, it could flip the whole car market, and not just Uber / Lyft, on its head. Imagine not just owning a car, but having it zip around making money for you as a robotaxi when you’re not using it—crazy IMHO! That’s where Tesla’s betting big, and it could mean we’re not just buying cars anymore but investing in little income machines. Not to forget, other companies like Waymo are in the game too, so it’s not just Tesla’s show. As this tech gets smoother and regulators give the green light, we might see a market where traditional car sales take a hit, and instead, it’s all about who’s got the best autonomous fleet. Prices could drop for rides, insurance might go nuts for non-self-driving cars, and cities could totally change with less parking and traffic - very few traffic lights too.

So, Warren Buffett’s been dumping stocks like Apple and Bank of America—$134 billion worth—and sitting on a massive $334 billion cash stash. Dude’s not spilling the beans in his annual letter, but it’s kinda wild, right? When the guy who’s all about buying great companies cheap can’t find anything worth grabbing, even with the market going nuts, it makes you wonder: are things getting too pricey? He’s still all in on equities long-term, but what’s he waiting for—some big crash or just a killer deal down the road?

While open-source LLMs offer transparency and community-driven innovation, the future might not be exclusively OSS. Proprietary models have significant advantages, including the ability to secure investment for cutting-edge development, customize for specific business needs, and maintain competitive edges through secrecy. Moreover, companies can directly monetize proprietary models, providing a clear path to profitability, and they can offer enhanced security and privacy controls crucial for sensitive applications. Thus, both open-source and proprietary LLMs are likely to continue playing vital roles in AI's future landscape.