AI products like ChatGPT much hyped but not much used, study says | Hacker News Reader

AI products like ChatGPT much hyped but not much used, study says

bbc.com

37 pointsYukidemama39 comments

From what I can tell, the biggest proponents of AI in everything are the investors anticipating that people will want AI in everything. I actually enjoy using LLM chat interfaces like ChatGPT directly, and I do benefit from Copilot in VS Code. I just treat it as a fallible entity just like I would a person. I use it as a sounding board or a place to jump-start ideas.

What I never use is anything resembling a product-specific AI interface within apps like Notion or Google Docs. I actively avoid AI writing assistants because I feel like authenticity is more detectable than some people realize.

I expect top innovations with with llms to be in other areas than bots and chats. There are just so many other things to automate first.

All I want is semantic search. I can make sense of what I find, I don’t want a Transformer to regurgitate slop to me

Right. I think there's value in being able to tell the LLM to give me the result in JSON format, for example. But the hallucinatory nature of LLMs poisons everything.

Yeah, I think I’m still not fully trusting in that pipeline because running it twice could give me different results, even if they’re structured.

Wish you could supply the random seed yourself to at least make it a little more deterministic

If you start digging into new topics lets say shaders or art history or whatever .. its hands down the best tool and teacher i ever had. You can get a broad overview over topics and repeat asking „stupid“ questions which are either hard to find on the net or you would not dare to ask anyone. I think these downplaying comments are kind of embarrassing for this kind of achievement in technology.

I'm curious about these "stupid" questions that are hard to find answers to on the net but an LLM can answer. Could you give an example?

Curious how you can even tell the llm result is correct when you are apparently unable to validate it with other methods

I am not unable to validate. Its just hard to get answers to specific questions. Once you have the answer and the corresponding vocabulary its not hard to find by web search.

For me the most significant help it gave was when learning shaders. Of course there is a large amount of tutorials but i am the type of person who has a hard time learning in a linear way if there a unclear points open (which perhaps are solved some chapters later). This type of conquering much ground by asking many questions is perfect for me and it was the first time i did not immediately gave up on shaders after two hours or so.

I am using ChatGPT every day. It helps me on my engineering tasks since it has a lot of knowledge. It also helps me on coding a lot but there I am using Github's Copilot. Certainly LLMs made my life way easier and they simplified many tasks that I was procrastinating and considering them huge hurdles on my everyday work and hobbies as well. This greatly improved my productivity and creativity! Trying to solve an engineering problem sometimes mind stucks in stupid ways to implement them. LLMs helped me to see another way of doing thigs greatly improving my efficiency!

So we have improvement on creativity, productivity and efficiency. What else do we need from LLMs really? I don't understand why people are not using it.

These will get transformed into agents, which will be like websites of yesteryears.

OK, so I'm neither a developer nor a content creator, and I can't think of anything useful a LLM could do for me. I tried a bit of chat and it was mildly amusing, but not for long. I certainly don't want one of them running my digital life for me. Isn't this just the zombie reincarnation of Clippy?

It certainly is if that's how you use it. It makes a fine Clippy replacement. But if you want to compare the views of Robert Hughes the art critic with the views of Carl Jung the psychologist and have the AI tell what they might say in agreement about AI, then it's going to give you a different experience.

> But if you want to compare the views of Robert Hughes the art critic with the views of Carl Jung the psychologist and have the AI tell what they might say in agreement about AI, then it's going to give you a different experience.

In my experience, trying to ask it a comparably detailed question on a topic in which I am fairly knowledgeable (and therefore capable of judging the response) results in output that is rarely insightful and not infrequently factually incorrect.

It's hit or miss depending on the topic. Marc Andreesson recently made a video about the mental space of AI contains low quality sections, average sections and super-genius sections. Claude Opus seems very well trained on art criticism and psychology. But even if it's not well trained, sometimes putting two things together that way can yield suprising results. An AI like Claude Opus makes this kind of knowledge surfing pleasant and fluid which is what you want when thinking creatively. Even the mistakes can spark insights because you have to ask yourself why you know it's a mistake.

But then this is just parroting, using statistical inferencing, what they "might" have said in their publications (and paraphrasing by countless academic papers, or even worse - since this is probably more likely - random qualified-or-not discourses and ramblings in internet forums), versus any sort of logical thinking, isn't it?

I doubt there are many acedemic articles or even internet forums comparing and contrasting the thinking of Robert Hughes and Carl Jung. A Google search of "Robert Hughes" "Carl Jung" turns up nothing except "Edward Robert Hughes" the painter. This ability of Opus to take a stab at it is what I find so fascinating. I agree, it's not a thinking reasoning being, but it's a pretty good mimic of one, which makes it more than a stochastic parrot to me. The natural language processing turns it to something relatable to a human as well. You might agree that a tool to do comparisons like this quickly would be useful for thinking:

Q: If art critic and Robert Hughes of "Shock of the New" and psychologist Carl Jung were to have a conversation about AI, what do you think they would they agree about it and where would they disagree?

Claude Opus: If art critic Robert Hughes and psychologist Carl Jung were to discuss AI, they might find some common ground, but they would also likely have some interesting points of disagreement based on their different perspectives and areas of expertise.

Potential areas of agreement:

Symbolic and archetypal representation: Jung's work on the collective unconscious and archetypes might resonate with Hughes' understanding of how art represents and communicates deep, shared human experiences. They might agree that AI-generated art could tap into these universal symbols and archetypes in powerful ways. The transformative power of technology: Both Hughes and Jung might recognize AI as a transformative technology with the potential to profoundly impact human consciousness and society, much like other significant technological developments throughout history.

The importance of creative expression: Jung's emphasis on the role of creativity in psychic health and individuation might align with Hughes' view of art as a vital form of human expression and a reflection of the human experience.

Potential areas of disagreement:

The nature of creativity: Jung might argue that true creativity stems from the unconscious mind and the integration of conscious and unconscious processes, while Hughes might place more emphasis on the role of conscious artistic intention and the influence of cultural and historical contexts on creative expression.

The value of AI-generated art: Hughes, as an art critic, might be more skeptical of the artistic merit of AI-generated art, seeing it as lacking the depth, originality, and emotional resonance of human-created art. Jung, on the other hand, might be more open to the psychological and symbolic significance of AI-generated art, regardless of its origins.

The impact on the human psyche: Jung might express concern about the potential negative impact of AI on the human psyche, such as the risk of over-reliance on technology leading to a disconnection from the unconscious mind and a loss of individuality. Hughes, while acknowledging the potential downsides, might be more focused on the ways in which AI could expand the boundaries of artistic expression and human perception.

The role of the individual artist: Hughes might emphasize the importance of the individual artist's vision, skill, and creativity in the creation of meaningful art, while Jung might place more emphasis on the collective and universal aspects of artistic expression.

Ultimately, a dialogue between Hughes and Jung on AI would likely be a rich and complex one, reflecting the multifaceted nature of the topic and the unique perspectives that each thinker would bring to the conversation.

Yes, that's interesting and somewhat amusing. But is it accurate or in any way beneficially worthy of billion dollar valuations??

If it's a new thinking tool that helps people solve problems better and quicker then it could be worth many billions and more

If that were some law of the universe wikipedia wouldn’t be begging for money regularly

Who said anything about a law of the Universe. The spreadsheet was the right tool at the right time so it took off. It didn't have to. People had calculators. But it so easy and useful. AI is the right tool for our time.

Do you have a business where you employ people to do office work?

It'll be embedded in applications in ways you won't even notice.

Mobile phone typing completions are suggested by AI models similar to LLM/ChatGPT, but using older and inferior architectures. These will be replaced with "Large" Language Models where the "Large" is a few hundred megabytes or a gigabyte at most. It'll be noticeably better, but not a "Chat bot".

Assistants like Siri will be completely revamped around text+audio models similar to ChatGPT 4o. It'll be a huge improvement that most people will notice and utilise daily, but it'll feel like a "Siri upgrade" not a new ChatGPT app.

Office apps will get AI assistants that will be like Clippy but actually useful. Excel and similar "more technical" apps especially would be much more useful to a larger group if the app held their hand and converted english queries into formulas for them.

Translation apps of all sorts will go from "usable but a bit meh" to nearly perfect.

Etc...

There's too much focus on consumer facing products. Where this technology really shines is behind the scened with NLP tasks. Imo, it's a bit of a weird perk I can have a conversation with my NLP.

Many people who are doing quite well with LLMs for whom they save many hours per week might not be fully forthcoming about their competitive advantage, especially with the controversy around these tools.

In the early days of the World Wide Web many people saw no need for a website, it’s still very early.

That's what NFT people keeep saying, too

Sports card collectors too

No shit Sherlock.

Until the end of the year we'll see both NVidia and the hyperscalers that rent their h100s see their revenues already shrinking, there's no way we can have that many companies doing LLMs and being useful.

I have no skin in this game myself, nor do I have any strong opinions one way or the other (though I too have a hard time imagining truly useful LLM-based solutions), but if you're willing to speak with such conviction, I'd like to see your short positions against NVidia et all.

To your last point, I’d like to point out there’s an adage that goes along the lines of “the market can stay irrational longer than you can stay solvent…”

Very true, which is part of why I don't have any positions on anything myself. But OP had a timeframe of 'until the end of the year'. If I felt strongly about something and had a timeframe like that, I'd place some bets.

But why would you expect someone else to behave like you in this tiny betting niche?

Good question. I guess I don't have any expectation that someone would. I was just stuck by the somewhat aggressive tone and strength of their statement. When people talk so aggressively I prefer to see them putting their money where their mouth is.

The other user has commented already, I don't want to bet against this craziness, and in a way, I'm certain that NVidia will continue moving up, until it isn't!

This is kind of a silly argument. Let's say, to take an extreme, you _know_ a stock will fall dramatically at some point in the next five years, but you don't know _when_. You know that, until it falls, it'll be pretty volatile, and sometimes jump up. So, when do you short it? You probably do not short it on day one; that's going to be horrendously expensive.

Really, betting against any tremendously hyped stock is a very dangerous exercise. Just ask those who shorted GameStop (which is _clearly_ dramatically overvalued, but is back to happily jumping up 10% in a day just because).

Perhaps more abuse than use.

Don't assume people are being honest about their use of ChatGPT...

Reply on news.ycombinator.com