Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

The force-feeding of AI features on an unwilling public

honest-broker.com

Post reply on HN
Print view View original

151–160 of 421 posts

Re: The force-feeding of AI features on an unwilling public

#151

Post by hatefulmoron »

Gigachad wrote: post #59
einrealist wrote: post #44

Earlier quoted context omitted.

If those smaller models are sufficient for your use cases, go for it. But for how much longer will companies release smaller models for free? They invested so much. They have to recoup that money. Much will depend on investor pressure and the financial environment (tax deductions etc). Open Source endeavors will have a hard time to bear the resources to train models that are competitive. Maybe we will see larger coop…

Seems like you don’t have to train from scratch. You can just distil a new model off an existing one by just buying api credits to copy the model.

"Just" is doing a lot of heavy lifting there. It definitely helps with getting data but actually training your model would be very capital intensive, ignoring the cost of paying for those outputs you're training on.

Re: The force-feeding of AI features on an unwilling public

#152

Post by bsenftner »

o11c wrote: post #134
oceanplexian wrote: post #91

Earlier quoted context omitted.

> LLMs are not very predictable. And that's not just true for the output. If you run an open source model from the same seed on the same hardware they are completely deterministic. It will spit out the same answer every time. So it’s not an issue with the technology and there’s nothing stopping you from writing repeatable prompts and promoting techniques.

By "unpredictability", we mean that AIs will return completely different results if a single word is changed to a close synonym, or an adverb or prepositional phrase is moved to a semantically identical location, etc. Very often this simple change will move you from "get the correct answer 90% of the time" (about the best that AIs can do) to "get the correct answer Whenever people talk about "prompt engineering", the…

What you're describing is specifically the subtle nature of LLMs I'm pointing at; that changing of a single word to a close synonym is meaningful. Why and how they are meaningful gets pushback from the developer community, they somehow do not see this as being a topic, a point of engineering proficiency. It is, but requires an understanding of how LLMs encode and retrieve data.

The reason changing one word in a prompt to a close synonym changes the reply is because it is the specific words used in a series that is how information is embedded and recovered by LLMs. The 'in a series' aspect is subtle and important. The same topic is in the LLM multiple times, with different levels of treatment from casual to academic. Each treatment from casual to formal uses different words, similar words, but different and that difference is very meaningful. That difference is how seriously the information is being handled. The use of one term versus another term causes a prompt to index into one treatment of the subject versus another. The more formal the terms used, meaning the synonyms used by experts of that area of knowledge, generate the more accurate replies. While the close synonyms generate replies from outsiders of that knowledge, those not using the same phrases as those with the most expertise, the phrases used by those perhaps trying to understand but do not yet?

It is not randomly changing things in one's prompts at all. It's understanding the knowledge space one is prompting within such that the prompts generate accurate replies. This requires knowing the knowledge space one prompts within, so one knows the correct formal terms that unlock accurate replies. Plus, knowing that area, one is in a better position to identify hallucination.

Re: The force-feeding of AI features on an unwilling public

#153

Post by arexxbifs »

brookst wrote: post #102
cs702 wrote: post #95

Your may agree or disagree with the OP, but this passage is spot-on: "I don’t want AI customer service—but I don’t get a choice. I don’t want AI responses to my Google searches—but I don’t get a choice. I don’t want AI integrated into my software—but I don’t get a choice. I don’t want AI sending me emails—but I don’t get a choice. I don’t want AI music on Spotify—but I don’t get a choice. I don’t want AI books on Ama…

It’s not spot on. Buying and using all of these products is a choice. The last is especially egregious. I don’t want poorly-written (by my standards) books cluttering up bookstores, but all my life I’ve walked into bookstores and found my favorite genres have lots of books I’m not interested in. Do I have some kind of right to have stores only stock products that I want? The whole thing is just so damn entitled . If…

Opting out is easy, we can just stop using products from Microsoft, Apple, Meta and Google. Of course, for many that also means opting out of their job, which is a great way to opt out of a home, a family, healthcare, dental care and luxuries like food.

I don't think it's entitlement to make a well-mannered complaint about how little choice we actually have when it comes to the whims of the tech giants.

Re: The force-feeding of AI features on an unwilling public

#154

Post by Paradigma11 »

satyrun wrote: post #85
daishi55 wrote: post #62

ChatGPT is the 5th most-visited website on the planet and growing quickly. that’s one of many popular products. Hardly call that unwilling. I bet only something like 8% of Instagram users say they would pay for it. Are we to take this to mean that Instagram is an unpopular product that is rbi g forced on an unwilling public?

My 75 year old father uses Claude instead of google now for basically any search function. All the anti-AI people I know are in their 30s. I think there are many in this age group that got use to nothing changing and are wishing it to stay that way.

A friend of mine is a 65 years old philosopher who uses it to translate ancient greek texts or generate arguments between specific philosophers.

Re: The force-feeding of AI features on an unwilling public

#155

Post by ToucanLoucan »

dang wrote: post #115

I also find these features annoying and useless and wish they would go away. But that's not because LLMs are useless, nor because the public isn't using them (as daishi55 pointed out here: https://news.ycombinator.com/item?id=44479578 ) It's because the integrations with existing products are arbitrary and poorly thought through, the same way that software imposed by executive fiat in BigCo offices for trend-chasing…

> It's because the integrations with existing products are arbitrary and poorly thought through, the same way that software imposed by executive fiat in BigCo offices for trend-chasing reasons has always been.

It's just rent-seeking. Nobody wants to actually build products for market anymore; it's a long process with a lot of risk behind it, and there's a chance you won't make shit for actual profit. If however you can create a "do anything" product that can be integrated with huge software suites, you can make a LOT of money and take a lot of mind-share without really lifting a finger. That's been my read on the "AI Industry" for a long time.

And to be clear, the integration part is the only part they give a shit about. Arguably especially for AI, since operating the product is so expensive compared to the vast majority of startups trying to scale. Serving JPEGs was never nearly as expensive for Instagram as responding to ChatGPT inquiries is for OpenAI, so they have every reason to diminish the number coming their way. Being the hip new tech that every CEO needs to ram into their product, irrespective of it does... well, anything useful, while also being so frustrating or obtuse for users to actually want to use, is arguably an incredibly good needle to thread, if they can manage it.

And the best part is, if OpenAI's products do actually do what they say on the tin, there's a good chance many lower rungs of employment will be replaced with their stupid chatbots, again irrespective of whether or not they actually do the job. Businesses run on "good enough." So it's great, if OpenAI fails, we get tons of useless tech injected into software products already creaking under the weight of so much bullhockety, and if they succeed, huge swaths of employees will be let go from entry level jobs, flooding the market, cratering the salary of entire categories of professions, and you'll never be able to get a fucking problem resolved with a startup company again. Not that you probably could anyway but it'll be even more frustrating.

And either way, all the people responsible for making all your technology worse every day will continue to get richer.

Re: The force-feeding of AI features on an unwilling public

#157

Post by queenkjuul »

brookst wrote: post #114
ikr678 wrote: post #111

Earlier quoted context omitted.

For consumer pproducts, sure, don't buy them. For people in office based careers, they may not get a choice when their company rolls out copilot, or management decide to buy an ai helpdesk agent, or a vendor pushes ai slop into the next enterprise software version.

How is that different from not liking other technology choices one’s employer makes? I could write a book about how much I hate our expense tool. But it’s never occurred to me that I am entitled to have a different one.

You should consider that yes, maybe you are entitled to a better one

Re: The force-feeding of AI features on an unwilling public

#158

Post by Xss3 »

CuriouslyC wrote: post #148
dang wrote: post #115

I also find these features annoying and useless and wish they would go away. But that's not because LLMs are useless, nor because the public isn't using them (as daishi55 pointed out here: https://news.ycombinator.com/item?id=44479578 ) It's because the integrations with existing products are arbitrary and poorly thought through, the same way that software imposed by executive fiat in BigCo offices for trend-chasing…

I am a huge AI supporter, and use it extensively for coding, writing and most of my decision making processes, and I agree with you. The AI features in non-AI-first apps tend to be awkward bolt-ons, poorly thought out and using low quality models to save money. I don't want shitty bolt-ons, I want to be able to give chatgtp/claude/gemini frontier models the ability to access my application data and make api calls for…

[flagged]

Re: The force-feeding of AI features on an unwilling public

#159

Post by greenavocado »

kgeist wrote: post #36
einrealist wrote: post #26

The major AI gatekeepers, with their powerful models, are already experiencing capacity and scale issues. This won't change unless the underlying technology (LLMs) undergoes a fundamental shift. As more and more things become AI-enabled, how dependent will we be on these gatekeepers and their computing capacity? And how much will they charge us for prioritised access to these resources? And we haven't really gotten t…

We run our own LLM server at the office for a month now, as an experiment (for privacy/infosec reasons), and a single RTX 5090 is enough to serve 50 people for occasional use. We run Qwen3 32b which in some benchmarks is equivalent to GPT 4.1-mini or Gemini 2.5 Flash. The GPU allows 2 concurrent requests at the same time with 32k context each and 60 tok/s. At first I was skeptical a single GPU would be enough, but it…

Qwen3 isn't good enough for programming. You need at least Deepseek V3.

Re: The force-feeding of AI features on an unwilling public

#160

Post by threeducks »

einrealist wrote: post #89
Gigachad wrote: post #59

Earlier quoted context omitted.

Seems like you don’t have to train from scratch. You can just distil a new model off an existing one by just buying api credits to copy the model.

Your "API credits" don't buy the model. You just buy some resource to use the model that is running somewhere else.

What the parent poster means is that you can use the API to generate many question/answer pairs on which you then train your own model. For a more detailed explanation of this and other related methods, I can recommend this paper: https://arxiv.org/pdf/2402.13116
Post reply on HN