No, "hallucinations" are in fact a deliberate outcome of their design and represent the LLM working exactly as it was built to. "Hallucinations" is a bad word to use because it erroneously anthropomorphizes the algorithm. It's an intentional marketing gimmick meant to trick people into thinking AI is more sophisticated and magical than it really is and condition users to excuse its "mistakes" as temporary mishaps or glitches that can be overcome, rather than seeing them for the hard limitations of the technology that they actually are.

That last part is essential to companies like OpenAI, because their marketability depends on your belief that ChatGPT is supposed to give you the right answers rather than the most likely answers. No amount of hard-coded guardrails and benchmark tuning (smoke and mirrors, respectively) will ever change the fundamental nature of its design.

Toward what–a cliff? What revolutionary breakthrough does Mr. Jassy think AI is going to bring to the table to offset $100B in capex investments? A slightly more sophisticated chatbot that can now correctly guess how many R's are in "strawberry" more than 80% of the time? ChatGPT has been around for several years now, and its "improvements" are both underwhelming and being delivered at a snail's pace with quite literally no promise of anything new other than a vague, undefined notion of "AGI" being right around the corner. There are no undiscovered use cases left for LLMs that are going to suddenly cause untold trillions of dollars in revenue to materialize out of thin air, and even if we boldly assume with no precedent whatsoever that AI will eventually evolve into some form that does enable that kind of revenue generation, we also have to assume the exorbitant cost of such a thing would completely nullify those dollars anyway. Beyond that, what "customer experiences" can Amazon hope to deliver via AI that are going to cause me to increase my consumer spending at a time when most people have to cut back? When does any of this even start to approach a threshold of making sense?

Until someone comes up with a use case that can generate billions in revenue to offset the insane cost of running these models at scale, it doesn't matter who "wins" in a race to the bottom.

SoftBank has already had to borrow tens of billions just to commit to their current pledges for Stargate. The idea that they can pledge a trillion is so absurd that even if you read it as satire, it would still be too unbelievable to be entertaining.

The image generator only "knows" what a style is by virtue of the training data being tagged as such, and while it's obvious that copyrighted material is included in this, you could theoretically have a set of source images based entirely on human-produced style imitators. For example, there is more than enough fan-made South Park or Simpsons style content on the internet that you could conceivably train a model to produce an adequate "South Park style" or "Simpsons style" image without anything that infringes on the actual copyright owner's material. What then? Even in this case, what does it say that you can still only reasonably get your desired output by specifically invoking the trademark in your prompt?

I think people are too focused on trying to frame the moral dilemma of AI image generation into the framework of existing copyright law despite it clearly not fitting, and this is highly unproductive IMO.

Nowhere in this does he actually say how that's going to happen. These tools have been around for several years now and have yet to do anything special except constantly hemorrhage billions of dollars, all while they beg consumers to find new use cases for their tools because they can't seem to come up with any themselves. The most these grifters can do is point to some nebulous promise of AGI just around the corner while being unable to define what AGI is and having no tangible roadmap for achieving it.

You can't say we're in some global AI arms race that we absolutely must win, and then say the best thing to do is for all sides to show each other their schematics. It's clear he doesn't know what open source actually means, because his logic only works in the context that China wouldn't have access to it, and DeepSeek has already proven that this isn't a barrier for them anyway. The fact that they're letting us copy their homework should probably signal to Eric that this isn't the arms race he seems to think it is.

This whole argument is just abhorrent nationalist fearmongering to insist that the entire scientific research community will somehow come to a complete standstill if they don't have access to cheap, red-white-and-blue branded AI; and really, what could be more American than subverting the gears of capitalism by insisting competitors work together toward a singular, national interest?

Maybe Eric Schmidt should have tried using AI to generate a coherent viewpoint.

You (and many others) are making a great case for the usefulness of AM radio, which is why I have a hand crank radio in my glove compartment despite my car still being outfitted with AM/FM.

You are not making a good case for the government demanding AM radio be built specifically into cars. If the goal is to ensure as many people as possible in as many scenarios as possible are equipped with AM radios in case of emergency, why not demand, for example, that all cell phones must be able to receive and play AM broadcasts as well? I'm well aware that's its own debate, but my point is that if any industry is going to be saddled with being responsible for everyone's emergency preparedness, I don't understand why they're targeting the automotive industry in particular except for the historical precedent of cars having radios built into them. If it was really a matter of practicality and/or public safety, the government should be expected to subsidize that cost, and it was simply a convenient coincidence that they didn't have to because cars already had radios in them; now, gradually, this is starting to change, and so this legislation comes across as more of a tantrum in response to the (still very distant) end of that free ride.

As a taxpayer, I'm all for public safety initiatives, but the NHTSA governs vehicle safety standards. Please correct me if I'm wrong, but I don't believe the inclusion of an AM radio is a matter of vehicular safety.

I'm not sure why I'm supposed to be impressed by 12% of a 33% growth in Azure (in its totality) at the cost of $20 Billion in CapEx, especially while OpenAI is still actively hemorrhaging billions annually. The optics of AI being "profitable" in the short term won't matter if they have to keep making these massive investments year over year while hiding the actual numbers inside of a product platform that would be considered entirely separate if not for Copilot being forcefully jammed into every aspect of it just so Microsoft can lie about it having a high adoption/usage rate.

I'm far more frustrated with the new trend of every drive thru playing some prerecorded message as soon as you pull up making it sound like you're talking to a person. I know well enough to just ignore it and wait, but I imagine the overwhelming majority of people immediately start rattling off their order only to be greeted with an extended silence or to be interrupted by the actual human, ultimately having to repeat themselves and unnecessarily extending the overall interaction time. I don't know what the intent was behind this practice, but it surely causes nothing but confusion and frustration for most people.

In the past few years, I've noticed a huge uptick in the proportion of applicants for virtually any type of IT position who now have some sort of certification or minor in cybersecurity on their resume, regardless of the nature of the position. This suggests whole cohorts of recent graduates have been told that cybersecurity is the "next big thing" and that they must get these certifications/minors to get a leg up on the job market, whether or not cybersecurity is actually their targeted profession. I would certainly never hire any of these people for an actual cybersecurity position, but I could definitely see smaller businesses which have identified a need for cybersecurity but can't afford to hire a true security professional picking up these workers. This must surely confuse statistics about positions filled/available, market rates, and even the very definitions of the terms involved.

There is an interesting similarity in the way the MSN article substitutes words (ersatz synonyms) to how machine translation between languages often makes the same errors.

Now if only we could do something about the "Please allow 7-14 days for us to process your request" nonsense when trying to unsubscribe from mailing lists. If you can sign me up (without my consent) the instant my address appears on some database somewhere, you can remove it just as quickly.

As someone who actually works in a school (grades 9-12), there is a prevailing mindset among teachers that kids "just get" technology and don't need to be taught anything. These teachers are mostly younger, tech-savvy millennials themselves, so it's not like we lack the capacity to actually teach this stuff, but no one sees it as necessary. I can only assume that the disconnect comes from our generation managing to figure things out on our own and expecting younger generations to do the same despite the technological landscape being very different. Many of our teachers expect incoming freshmen to already be fluent in software like Word and Excel because their curriculum depends on it, and they have no contingency plan for instances where this is not the case, leading to broader academic issues for these kids.

I am reminded of this article about college students, a worsening problem which obviously starts younger, but it seems like no one wants to step up and do anything besides complain about it. It's frustrating because this should be an easy problem to solve, but too many people are blind to there being a problem at all until it's too late.