FWIW I found a series of TikTok accounts with people teaching French in one way or another, with apparently automated subtitles. Quite useful although the subtitles are extremely tiny and quick to disappear from the screen. That said, an interesting way to hear lots of individuals on different topics, I ducked out of the one on Sartre but liked some talking about different slang or "native speaker" grammatical constructions. These seem a lot more useful than say trying to watch a movie with its impenetrable dialogue.

I'm curious and hopeful about this, if it could make it inexpensive to learn (skeptical about that part). However, I spent years in language labs learning Japanese with a Tandberg cassette tape recorder. It had a dual volume control so you could match your voice to the speaker's which was fabulous and was especially effective interspersed with summer work overseas, I remember one student in 4th year unfortunately ended up sounding like a tape recorder to us. Another student sounded like a charismatic NHK news announcer somehow! We were just kind of in awe. ;) I haven't used Duolingo and this might be a good learning tool, but I'd caution the student to combine it with talking to actual live humans in any way possible, lest you somehow learn a pronunciation, inflection or (shudder) hallucinated idiom from generative AI. I'd say I am fluent but even so, Japanese movies are difficult to understand for me so your mileage may vary. Well, I have trouble with English song lyrics too so could be a me problem. That said, I remember trying Plimsleur CDs for Spanish but they only had one that used a dialect calling the word for the language "castellano" instead of "español". I tried learning Chinese from phone app briefly, so if you like learning languages this could be a lot of fun. I would be very interested in having an app that could produce a subtitle overlay based on AI's understanding of the text as opposed to the shortened subtitles you often get. All these things sound great but also not sure if this will lose people jobs or create false "imagined" language artifacts that ickily make their way from genai into popular discourse.

Should be easy to add list of trusted repos, and check sigs if there is such a thing.

Didn't see it in TFA but it is probably talking about Coelux. You laugh but shit this looks really sweet.
Still want some UV and be able to crack a window open though.
One example of their products:
https://ancillary-proxy.atarimworker.io?url=https%3A%2F%2Fwww.coelux.com%2Fen%2Fp%2Fco...

It seems you would get less hallucinations and more reliability by taking the same fee and hiring a firm in a country with lower standard of living, unless quick replies are necessary. There are plenty of smart people and mistakes or half-assedness (like I have seen from a huge Indian marketing team a client has) could be ascribed to AI, but you would never be getting hallucinations just maybe misunderstandings or amateurishness. There are smart, quick people there, just maybe not as lightning-fast and sleep-agnostic as an LLM.

Guess what, even if your company somehow contracted with OpenAI to not steal and train on your secrets, now they can do so under the guise of targeted advertising or maybe employee monitoring. They can do both while also training and triple-dip! If they buy Chrome you might want to change your browser or your LLM provider in an attempt to limit introspection into your private affairs. I doubt it would work though since your cross-site activity data can be purchased regardless of your browser. Consider them selling a service to a special user tier that lets you ask LLM questions about a specific person and speculate on their future activities. It will be easy and one RAG away once this loop is closed.

As I posted above, based on TFA not a click-baiting alarmist blog but a site made by a documentary filmmaker with two lead-poisoned children. But do troll on.

The owner is not impartial since her children are lead-poisoned. But the site says it is science based and reproducible. I think in the case of lead and other heavy metal poisoning this is spot-on. You don't need to take the side of the heavy metal producers, lol! Snipped from the site:

Tamara Rubin is a multiple-federal-award-winning independent advocate for childhood Lead poisoning prevention and consumer goods safety, and a documentary filmmaker. She is also a mother of Lead-poisoned children (two of her four sons were acutely Lead-poisoned in 2005).

Tamara owns and runs Lead Safe Mama, LLC — a unique community collaborative woman-owned small business for childhood Lead poisoning prevention and consumer goods safety.
Since July 2022, the work of Lead Safe Mama, LLC has been responsible for six product recalls (FDA and CPSC).
All test results reported on this website are science-based, accurate, and replicable.

Now those out of work, unknown actors can get their names out in (badly acted) incomprehension and anger about how their unknown likenesses were used in shady advertising with similarly miniscule audiences. Suddenly their names are on international media! (Thanks /.) In other words some of them may be stunned but now they can google their own names and find something.

There are different aspects of "verification". You wouldn't trust the existing method with your bank account. Posting from a domain name is not proof against spoofing or other kinds of betrayal. Way back when (mid 90s) where I am, to get a corporate domain name you had to pay a lot of money to one of a few trusted security companies and supply paper documentation they would review. Because the verification authority's name was on the line. Might still be that way for some. Though I don't use bluesky yet I'd expect that a verification method and verifying authority both selectable by the user would be more inline with the ethos of bluesky. So identity verification could be a layer added on top and potentially based on a distributed mechanism as well.

What a f*ing disaster. Of course so many bigger disasters going on in the U.S. compared to little ol me's purchasing needs. But am overseas, meaning might even get another tariff to ship it here if they are assembling it in Cupertino. Need to replace my vintage MBP and wanted a near top of the line MBP. Now considering whether to go for a much cheaper model, but if there is no end in sight I might need this to last a while.. That and maybe I'll need to locally source a windows or linux machine that already is in a shop nearby!

...this kind of alien logic:
- Can only be done via AI "magic" that makes it plausible in his marketing pitch
- Proprietary prompt-driven "thinking" used so locks in xAI to get the SSA system contract forever
- Plans to kill SSA or maybe in the name of "efficiency" setting it to a fixed blanket payment per person that could be coded by a highschooler
- Brings with it 100% visibility into SSA data and everything connected to it
- Having the data means xAi can then set up a "secure" identity service that stores everyone's passwords and authenticates everybody to every app, I mean he knows your SS# and he'll get your bank info in some other similar way. And.. that is what the WeChat-style app that X should become would do. Everything, basically.
- While killing half of SSA he can take x% of its annual payouts as his cut, it's a per-head license fee (lol).
- Or, just replace with a private SSA owned by techbros.

Really there's nothing but upside! /s

If anyone has read the news recently you'd know we as a species have a miserable track record regarding ethics and in particular psychopathy, narcissism and sadism are in control of our societies. The problem is with power ethics and their policing goes out the window. Also, it was not so long ago that Sweden was reviled for embryo research. This is about 10,000x worse. Once the cat is out of the bag every country and technologically advanced corporation will have ample opportunity to reinterpret this without the veneer of ethics. At least until they learn how to grow the desired organs by themselves, or bodies without heads which sounds like iit won't work, they will simply grow normal bodies and kill them. This is the definition of a slippery slope. Put it this way, if it only took a decade for AI big money to change "do no evil" to "let's make autonomous hunter-seeker military AI drones", we will likely see a similar dystopian future in the biological sector.

...to think up adversarial challenges, test and check for them?
But what happens when LLM is trained on a repository that itself contains malicious code.
It sounds like we need adversarial input checks touching on all points of the food chain.

So I've been wondering how good is Apple's neural engine at calculating the transformers compared to Nvidia and is there a chance they could improve training by beefing it up i.e. adding more neural cores? Looking at buying an M4 64GB MBP and having to use cloud for training if needed. Is there a shred of a possibility that Apple could release coprocessor modules that could be added to the motherboard or even connected via USB cables, even if it is a box with Nvidia chips in it to improve their training performance? I mean some guys are doing stuff with Mac minis but I don't know if the neural cores are inside the M4 or are they separate chiplets that could be greatly multiplied.