Useful for who? Who is the customer that wants this? Especially when is output is untrustworthy?
I guess generating nonsensical bullshit is 'useful' to those who like sending spam emails and enhancing their scams across the internet. Makes it easier to not rely on 'GPT' search engines or silly use-cases like generated cover letters at all.
GPTs cannot reason or explain their own output. Even worse as it is often confident about giving the wrong answers. It is a another solution in search of a problem.
I get real value out of GPT-based models for software development. I've had ChatGPT answer questions about particular APIs I don't use often, and Copilot has made it faster to rapidly iterate on things because I don't have to type out boilerplate.
Early on I was pretty dismissive about both, and thought they were at best a crutch for junior engineers. But I've come to appreciate that they have the capacity to remember many more APIs and idioms than I do, so when I'm venturing outside of my usual languages/APIs, that's where they really provide value.
As SE you can use it as a dedicated junior doing tedious snippets and small programs for you. Not always a hit but often right or at least outline it right is enough to collect value.
And with Google getting worst results by the day and often not finding anything, these feel like a step ahead.
The math behind subprime mortgages was also kind of "solid" (something about stochastic PDEs). Math can be "solid", but disconnected from the reality.
(The connection to reality is impossible to prove mathematically).
Whenever I have difficulty finding the info by googling, I go to chatGPT and ... it gives me the wrong answer with confidence and aplomb. It is never in doubt.
(Eventually, the info can be found by more googling).
By now I am collecting the examples of ChatGPT and now bing getting obvious things spectacularly wrong, and sharing them with the less techy friends, as a warning.
That said, I treat the GPT answers the same way I would treat a “question dump” for a certification: it’s almost certainly wrong, but it gives me an idea where to dig further.
Even incoherent babbling with the right terms uttered can provide a useful set of anchors for future learning…
This indeed often works, but there's a whole class of questions where it doesn't.
Example: yesterday, I was listening to Abbey Lincoln's performance of the song "Angel Face" (you can find it on youtube). I wondered who composed such a beautiful song. Unlike the majority of jazz standards, this song doesn't have a dedicated page in wikipedia. Other sites cited different composers. When I asked chatGPT, it confidently told me the song was composed by Duke Jordan in 1952.
This claim was easy to refute by googling Duke Jordan. By more googling, discovered the real composer: Hank Jones (1947, originally as an instrumental piece).
This is an instance representing the class of questions where you expect a very concrete answer, but chatGPT fails, and the info it provides is totally useless.
I noticed that ChatGPT is absolutely terrible at many “factual” questions, where the answer is a date or person or list of such.
On the other hand it seems better at more abstract questions, as well as giving interesting background around some of the “why?” ones.
The coding ones I would limit to very short tasks that are easy to eyeball and verify: on one hand, the other day in 5-6 iterations it wrote me a functional prototype of an audio player (because I was bored to try to piece together the complexity that is WebAudio api); on the other hand once I tried to ask it to write a file upload code for a Rust web server ride - and it came up with three plausibly looking but totally invented frameworks :-)
So, I won’t let it do anything in an unsupervised fashion.
the longer I observe the VC market the stronger I believe what you said is true. The goal for VCs is to get to the IPO to sell hyped shares to retail investors.
Keynesian beauty contest [1]. They don't pick the best assets. They pick the assets that they think most people are going to think are the best assets.
Wouldn't a Keynesian beauty contest be more like they don't pick their personally favorite assets, they pick the ones that they think are going to perform well in the market? As a kind of strategy?
A pre-seed investor invests in assets they think seed investors are going to like, seed investors invest in assets they think series-A investors are going to like, series-A investors invest in assets they think series-B investors are going to like, ..., series-whatever investors invest in assets they think IPO-subscribers are going to like, IPO-subscribers subscribe to assets they think the secondary market is going to like.
At some point in the process someone goes: ...hang on a minute. But by that time, the asset is owned by you and me, through the indirection of pension funds, sovereign wealth funds, and the like.
This is, I think, essential to any equity market. Some domains are so boring that only inside information rivals judgement. Cardboard boxes. Some other domains are so bubbly that feeling the zeitgeist of the crowd might outperform judgement. Tulips.
Recently I came to the same conclusion. It is the only one, that explains the utter lack of due diligence done in the last decade or so. From that perspective, it also makes absolute sense.
Nvidia announced their earnings Wednesday night, as well as some other news, and the stock jumped about $29 (about 14%) right after the stock market opened.
ChatGPT runs on Nvidia GPUs. Stable Diffusion was created on Nvidia GPUs. Nvidia said the higher than expected sales growth was due to sales to data centers, which seems more related to deep learning than crypto-mining and the like.
Martin Skreli (yes, pharma bro) did an analysis of Nvidia recently, highlighting the importance of the data centre revenue segment to Nvidia's business. Basically they need that segment to grow and grow over the coming years, otherwise there is no real justification for their sky high P/E (crypto mining is in the toilet, many gamers already upgraded their GPUs during the pandemic, possible recession looming so fanciful overpriced GPU spending will be impacted).
I’ve wanted to get into NVidia for a while as I’ve seen crypto, ai, big data being waves of interest. It’s very hard to judge an entry point though and the stock has been quite volatile.
And this is right as the economies of miniaturization are drying up. Nvidia's 40-series consumer cards have hardly any value improvement over the last generation. They're just higher priced on the same performance/cost scale. That kind of stagnation will obviously slow sales since the ROI of upgrades takes longer to go positive.
Problem is, lots of players in the ML world are making their own homegrown accelerator cards. Facebook, Google (TPU), Tesla (Dojo), etc.
I think the next decade will be one of fragmented and incompatible compute platforms, with all the big companies trying to do AI-as-a-service, rather than selling accelerator cards.
The poor man without a multi-billion company behind them will end up buying NVidia kit.
Crypto and layoffs are out boys! AI chatbots are back in!!
Edit: also since history is repeating itself, can anyone remind me what the hot investor craze was after AOLs Instant Messaging became a thing? Second life or something? Should I be buying shares of Roblox?
Yes, of course, Google would not be serving up an auto-complete for such a high volume search if had no demand and/or click through for the related search results; just take a look at the results yourself to see:
I don't understand that bar chart. It shows something like 11bn in investments, and then there's an asterisk and it says "includes 10bn by Microsoft", raising the interesting question: Is the set of investors going nuts for ChatGPT really broad-based, or is it just a few institutional-level players who happen to be whales when it comes to tech investment?
… but then the prices would - outside of occasional hype peaks - plummet back down to some realistic value instead of remaining at their surreal artificial high.
I mean, investors have never in the past pursued fads or trends or technology they didn't understand the first thing about. Quick, follow that invisible hand!
63 comments
[ 1.7 ms ] story [ 125 ms ] threadI guess generating nonsensical bullshit is 'useful' to those who like sending spam emails and enhancing their scams across the internet. Makes it easier to not rely on 'GPT' search engines or silly use-cases like generated cover letters at all.
GPTs cannot reason or explain their own output. Even worse as it is often confident about giving the wrong answers. It is a another solution in search of a problem.
Early on I was pretty dismissive about both, and thought they were at best a crutch for junior engineers. But I've come to appreciate that they have the capacity to remember many more APIs and idioms than I do, so when I'm venturing outside of my usual languages/APIs, that's where they really provide value.
As SE you can use it as a dedicated junior doing tedious snippets and small programs for you. Not always a hit but often right or at least outline it right is enough to collect value.
And with Google getting worst results by the day and often not finding anything, these feel like a step ahead.
Whenever I have difficulty finding the info by googling, I go to chatGPT and ... it gives me the wrong answer with confidence and aplomb. It is never in doubt. (Eventually, the info can be found by more googling).
That said, I treat the GPT answers the same way I would treat a “question dump” for a certification: it’s almost certainly wrong, but it gives me an idea where to dig further.
Even incoherent babbling with the right terms uttered can provide a useful set of anchors for future learning…
Example: yesterday, I was listening to Abbey Lincoln's performance of the song "Angel Face" (you can find it on youtube). I wondered who composed such a beautiful song. Unlike the majority of jazz standards, this song doesn't have a dedicated page in wikipedia. Other sites cited different composers. When I asked chatGPT, it confidently told me the song was composed by Duke Jordan in 1952. This claim was easy to refute by googling Duke Jordan. By more googling, discovered the real composer: Hank Jones (1947, originally as an instrumental piece).
This is an instance representing the class of questions where you expect a very concrete answer, but chatGPT fails, and the info it provides is totally useless.
On the other hand it seems better at more abstract questions, as well as giving interesting background around some of the “why?” ones.
The coding ones I would limit to very short tasks that are easy to eyeball and verify: on one hand, the other day in 5-6 iterations it wrote me a functional prototype of an audio player (because I was bored to try to piece together the complexity that is WebAudio api); on the other hand once I tried to ask it to write a file upload code for a Rust web server ride - and it came up with three plausibly looking but totally invented frameworks :-)
So, I won’t let it do anything in an unsupervised fashion.
[1] https://en.wikipedia.org/wiki/Keynesian_beauty_contest
A pre-seed investor invests in assets they think seed investors are going to like, seed investors invest in assets they think series-A investors are going to like, series-A investors invest in assets they think series-B investors are going to like, ..., series-whatever investors invest in assets they think IPO-subscribers are going to like, IPO-subscribers subscribe to assets they think the secondary market is going to like.
At some point in the process someone goes: ...hang on a minute. But by that time, the asset is owned by you and me, through the indirection of pension funds, sovereign wealth funds, and the like.
There, I said it.
The irony would be that it would learn everything they know and be just as terrible.
Exact replacement. POSIWID.
https://www.youtube.com/watch?v=USKD3vPD6ZA
https://www.google.com/finance/quote/NVDA:NASDAQ?window=5D
ChatGPT runs on Nvidia GPUs. Stable Diffusion was created on Nvidia GPUs. Nvidia said the higher than expected sales growth was due to sales to data centers, which seems more related to deep learning than crypto-mining and the like.
https://www.youtube.com/watch?v=nn5pWp6nzKM
I think the next decade will be one of fragmented and incompatible compute platforms, with all the big companies trying to do AI-as-a-service, rather than selling accelerator cards.
The poor man without a multi-billion company behind them will end up buying NVidia kit.
Edit: also since history is repeating itself, can anyone remind me what the hot investor craze was after AOLs Instant Messaging became a thing? Second life or something? Should I be buying shares of Roblox?
This time, we did that one before chatbots
At least the LLM stuff is useful. I haven't gotten this much value out of a new technology in a long, long time. Possibly since the Web.
https://google.com/search?q=chatgpt+stock
That’s part of the game…
Don't forget Excel, and even the original MS-DOS. In fact it would be hard to name a single successful Microsoft product that wasn't an acquisition.