> they'd rather vibe code themselves than trust an unproven engineering firm
You could cut the statement short here, and it would still be a reasonable position to take these days.
LLMs are still complex, sharp tools - despite their simple appearance and proteststions of both biggest fans and haters alike, the dominating factor for effectiveness of an LLM tool on a problem is still whether or not you're holding it wrong.
The food industry has pretty much invented the whole process of making "addictive" products and then "test[ing] out recipes on the public to make it even more addictive". Of course, we usually call it making products that taste good, and running taste panels with the public for product development (making a new tasty thing), quality control (ensuring the tasty thing stays tasty), and market research (discovering even tastier things to make in the future). Each part of it employs all kinds of specialists (and yes, those too - nutrition psychology is a thing).
The process is the same. The difference between "optimized for taste" and "addictive" isn't exactly clear-cut, at least not until someone starts adding heroin to the product (and of the two, it's not the software industry that's been routinely accused of it just for being too good at this job).
Not defending social media here in any way. Cause and effect is known these days, and in digital everything is faster and more pronounced. And ironically, I don't even agree with GP either! I think that individuals have much less agency than GP would like it, and at the same time, that social media is not some uniquely evil and uniquely strong way to abuse people, but closer to new superstimulus we're only starting to develop social immunity to.
I imagine that apostrophes in legal writing are trouble, much like commas. It's too easy to shift or even drop one them by mistake, which can alter the the meaning of the whose sentence/section in unfortunate ways.
Showing my age here, but the original samples are available too, and in MP3 or WAV format - they're in the installation directory of the game (in case of StarCraft and W3, hidden in a weird pseudo-ZIP data file (used to call it "Virtual File System")). That's where we sourced them from to set them as system sounds, back when Windows versions were still in four digits.
If you're enough of a fan to want to use these voices, chances are you still have the original installation media (or original bootleg copy) somewhere around the house :).
Now I'm still waiting for someone to succeed at a clean-room recreation of Majel Barrett's voice, so we can finally have computers sound like they always should have.
We could've been there a decade ago, but the high-quality audio samples, made officially and specifically with possibility of this use in mind, got trapped somewhere between the estate, producers, and a commercial interest that called dibs, and then procrastinated on the project instead.
I did this. She recorded clean (imo, i cleaned it up) audio for “Star Trek: The Next Generation Interactive Technical Manual” which is available on archive.org.
Nurse Chapel, and "Number One"* from the original series' original pilot, The Cage. Both of these characters are main cast in SNW, sadly no mind-swap plot with these two has happened yet.
* I don't think she had a full name at that point?
I just yeeted a bunch of extremely noisy fragments into elevenlabs, and it came out pretty good on their cheap $5 plan. If you're after this for your own amusement, let me know if you want a screencap, or a dump of the source files.
Obv no clean room reconstruction but good enough for personal use...
I have lots of super high quality, clean audio recordings from her ripped from an old video game that she did voice work for. I've tried various TTS models over the years with it. Getting the pitch and tune is easy, but getting the impersonal detached robot-y feeling is kinda tricky. But I haven't tried in the past 6 months, so maybe it's time to give it another shot.
the inflection and impersonal feel is definitely hard to get right. there are parameters in the elevenlabs API docs to make the voice more stable (= monotonous; see speak.sh in that repo) but still the voice cloner on my $5 plan doesn't really get it right.
nevertheless... i'm still having a lot of fun with this.
edit: if I am forced to rot my brain with the 10x productivity boosting slop gun, at least I'll do it grinning
> pod cleaned up. waiting on the behemoth to finish grinding through Italy.
< if only postgres had progress indicators
... then they coulda called it progresql
> lmaooo
> Bash(~/speak.sh "Joke detected. Humor subroutine engaged. Ha. Ha. Ha.")
“Director John Badham states in the commentary that the actor voicing the raw content that was later modified for the computerized effect was John Wood (the Falken character), reading the script word-for-word in reverse order in order to portray a "flat quality" with limited inflection. That raw audio was then edited and re-assembled after being run through audio processing equipment to achieve the desired effect.”
Apparently John Wood read the lines in reverse order to make the enunciation weird. If you train a model, feed the lines you want in reverse word order, then split on silence and reverse them again, you should come close.
> if OpenAI can essentially work with Monopolly money, whey can´t "we" do it too?
The answer is, in case anyone wonders: because OpenAI is providing a general purpose tool that has potential to subsume most of the software industry; "We" are merely setting up toll gates around what will ultimately become a bunch of tools for LLM, and trying to pass it off as a "product".
For the huge amounts of capital already burnt, and another 1T in CapEX RPO being announced last few weeks and months, isnt that too many of "potentially" and "ultimately" unspecific qualifiers you are throwing around here? Reminds a lot of Sam Altmans classic lines of unspecific statements like "Codex is so good" or "I can only imagine how good it will get" by the end of 202x (insert year of the decade according to your own preference). After 10+ years of OpenAI and 4+ years of ChatGPT, why is the potential not materialising ?
We are in an era of solutions being invented and developed at a rapid pace. The identification or invention of problems that they are good at solving has a much slower lifecycle.
Counter argument just to play devil’s advocate. Is that forming LLMs into useful shapes could become the game. If it turns out to be impossible to build a real moat around making LLMs - like maybe China or just anyone will ultimately be able to run them locally / cheaply, then the game of spending a billion dollars training one is much more risky
> Not to mention collecting them at all means those servers are a primo location for state actors to stage themselves to make copies of data before being deleted.
Not to nitpick, but in this case they'd be collecting data they already own.
For state actors - they frequently have issues "connecting the dots". Or heck - maybe connecting the dots is easy but it's a manual process that introduces too much friction for them to do casually. Maybe some of the data they connect it with is not trustworthy.
If the dots already come pre-connected, it makes the job easier.
Not to mention its value as blackmail material shoots up because it comes pre-associated with your government ID and/or a scan of your face because fewer sources/methods need to be risked.
In addition to the sibling comments, even if they do own the ID itself, they do not own the association with Discord users, and the ID might also be faked.
> An agent that can truly “use your computer” is incredibly powerful, but it's also the first time the system has to act as you, not just for you. That shifts the problem from product design to permission, auditability, and undoability.
Or rather, just reveals that the industry never bothered to properly implement delegation of authority in operating systems and applications, opting instead to first guilt-trip people for sharing their passwords, and later inventing solutions that make it near-impossible to just casually let someone do something for you.
Contrast with how things in real life function, whether at family level or at the workplace.
Half-agree. As an industry, we do suffer from not-invented-here syndrome, so have a lot of mediocre attempts to implement it that don't learn enough lessons from historical human-human examples.
Delegation can also be scary with other humans. "Power of attorney" and all that. Or even just micro-management.
I can sort of understand it if I squint: every feature is a maintenance burden, and a risk of looking bad in front of users when you break or remove it, even if those users didn't use this feature. It's really a burden to be avoided when the point of your product is to grow its user base, not to actually be useful. Which explains why even Fischer-Price toys look more feature-ful and ergonomic than most new software products.
You could cut the statement short here, and it would still be a reasonable position to take these days.
LLMs are still complex, sharp tools - despite their simple appearance and proteststions of both biggest fans and haters alike, the dominating factor for effectiveness of an LLM tool on a problem is still whether or not you're holding it wrong.
reply