I noticed the scarcity a bit in the last few months. I have been giving them to my dog for years for the low mercury omega 3s -- I even had a veterinarian stamp of approval.
While there are bad engineers, I imagine that teams that have evolved alongside AI and have maintained their size and consistency without adjustment may have structural issues. I would argue that team size counts ought to be smaller now than before Claude. Unfortunately this idea could feed an AI-induced layoff trend, but it could instead lead to product diversification instead.
It absolutely does not absolve them of sin. But let's give them credit where it is due. Open models are a really good thing. Hopefully I won't be lynched by a mob in 3 years for once saying this.
While Zuck's contributions are probably overwhelmingly more negative than positive, Meta initiated both pytorch and llama which have had profoundly positive effects on AI research. And I really am not a fan of python :D
ごちそうさまでした。「gochisousama deshita」 This can be said to the cook. I heard "いただきます！"「itadakimasu」more frequently for a variety of reasons. But it is common to shout out to thank the cook when you leave a restaurant. I became comfortable doing it myself.
Money is deep on both "sides". It didn't help Harris to have the most either. But you can't deny that some political steerage toward donor agendas occurs.
Caps Lock is such a tragic throwback. Even after decades of ridicule "Caps lock is cruise control for cool" keyboard designers don't get that we don't want it.
I actually like the number font, the presence, but the loss of visibility for the shifted symbols is a huge loss. I know where the numbers are -- I admittedly need more reminding for the symbols as I never explicitly practiced them. The muscle memory for them isn't quite as strong and in low light I need to see them.
Cramming the fn key into the sacred ctrl-alt-cmd triumvirate is disagreeable. Would require learning and remapping. I prefer fn to be distant as I have chorded for decades with the said trio.
I live in an American city. I have a yard large enough to play frisbee with my dog. I walk downstairs to go to work. I can walk to a good taqueria and take pole dancing classes a few doors up (someday). There is an evil bank that I could walk to but don't. Cafe, Thai, Soul Food, Vietnamese, Pizza. Closer than all of the walking distance I mentioned is light rail that goes downtown and further. Someday it might even go to Vancouver (the other one sadly :( )
It sickens me when people use the word plagiarism to describe LLMs. Particularly when they use words and language that they themselves have plagiarized -- words like "love" and "economic" -- where do you draw the line? No, we haven't figured out how to fairly compensate everyone -- the frontiers aren't even trying. Spotify does at least try in their domain (and fails badly). Regardless of how the entitled get paid, the plagiarism label isn't fair and is diametrically opposed to Fair Use. Fair Use allows the commons to build on its history. Those against are typically parroting IP conglomerate conservatism and localized greed. I see immense benefit from the distillation of all available knowledge with an LLM interface. A recent article posted here says it perfectly "AI rewards expertise". I see vast benefits -- am willing to pay a significant monthly fee -- and sense that many are simply blind to and inarticulate about the growing force multiplier potential of modern AI.
I don't think the open source models are anywhere near ("we’re talking around four months of lead time") Anthropic's Fable for long-horizon context planning. Sol isn't either. The benchmarks are quite misleading. The short-term question is how many customers need these deep problem solving capabilities? The long-term question is what will the frontier labs build (deepening and broadening of capabilities and products) if they are capitalized as they desire?
This article is revisionist and lost -- it decidedly disagrees with my direct memory. I had Snow Leopard on a vm for years -- not because I was misguided, but because it was a very stable release in many dimensions and deserved its status.
I am much later in my career, and I have established many skills that have atrophied which I have restored in need and/or desire over a span of decades. I think it is fair for younger people to be much more cautious with AI assisted coding tools (e.g. Claude Code, Codex) and their impact upon their skills. But understand that some atrophy may not be bad and simply a symptom of focus. If you want to stay a master juggler, keep juggling. But you can also ask yourself, do I need to be master juggler? You will still be able to juggle regardless.
I have not worked on a project with 20 AI-assisted contributors. I have worked on a team with 4. While smaller, I can already see that the code churn, and conflict resolution is more a kin to a larger team. However, we also have been introspective about how we can interoperate with these new disruptive tools -- ritualizing the LLM's ability to create and maintain documentation and perform code reviews has been indispensable. Perhaps 20 on a project is unsustainable, and even unnecessary.
I was hand-rolling NEON SIMD 15 years ago, and in many cases the compiler (clang/llvm) simply out optimized me. I kept the attempts that were better than what the compiler could already do. That was ARM NEON, quite new at the time, not SSE, and again that was 15 years ago that the compiler could already beat me much of the time. I hate the naive cult coder adage concerning premature optimization, but this might be a situation where you peruse your compiler output before you start writing code in a manner the compiler can for you. In Clang, auto-vectorization is enabled by default at optimization levels -O2 and -O3.
The cumbersome and poorly designed workflows surrounding passkeys are a significant hindrance to their adoption as well. It surprised me how bad they were when they emerged -- it was as if the UX was intentionally designed to kill passkey adoption.
The "race" has multi-dimensional impacts. This story parallels only some of them. "Nobody wins from the race" completely ignores the generality of AI. Xi Jinping highlighted this week that he clearly understands this multi-dimensionality; your words do not.
A and B examples are convincingly "human". It is complete nonsense to say otherwise. The AI versions are simply giving the music a kind of commercial streamline production treatment; treatment that many listeners need to accept the music as viable. Some will fight brutally to hide this need here. No offense, but both a and b are thoroughly cliché. The AI is not sucking away some grand uniqueness of expression here. I have heard both raw and produced countless times. Humans are generative AIs already -- fighting machines that do it is ironic and dishonest. Sure the AI process may be robbing some of the "honesty" of the performance. But you guys aren't unique like Daniel Johnston when you are unfiltered so I don't see a problem with either version. The studio is a place of magic and sonic makeup. If you want to wear makeup, go for it.
OpenAI offered ChatGPT to the world. A large, monied cross-section of the world had yet to throw its capital behind the Large Language Model technology that made the ChatBot possible. While it is fair to see AI development now as a global imposition, OpenAI did not have the agency as a 2022 startup to impose on the scale we see now.
I have flirted with the idea of returning to Slashdot, as I have noticed a few examples of suspect conflict-of-interest moderation occurring here on HN, given its y-combinator hosting; particularly with respect to posts related to prediction markets.
For typical listening (though humans can perceive bone-conducted vibrations up to 100 kHz or even 120 kHz) 16-bit-fixed/44.1kHz is a high-fidelity transport format. As a DSP researcher, I prefer 32-bit-float/44.1kHz as a transport format. I often upsample to 32-bit-float/188.2kHz or even 32-bit-float/192kHz for signal processing applications such as high-fidelity reverberation via direct and FFT convolution. While the author advocates for the transport to ear use case, I would argue that 24-bit/192kHz provides greater fidelity and resolution for sound processing. I found the pedantic arrogance of the author to be annoying. But yes, the sampling theory is an important consideration -- but so is the quality of the actual digital filters used in the DAC->ADC pipeline. They are much more forgiving and less lossy at 192kHz.
"An LLM predicts the next word based on all the words before it. That is the whole story. There is no idea sitting underneath." At most, and I will argue for less, that is half the story. This egregious over-simplification doesn't account for training and the massive dimensional space of thought it forms. Given that frontier models likely have trillions of parameters, the combinatorial space of ideas within approaches the infinite -- hardly zero.