O’Reilly urges world to get ready for a red-hot summer of AI | Hacker News Reader

O’Reilly urges world to get ready for a red-hot summer of AI

aimagazine.com

32 pointsrbanffy16 comments

I'm not very interested in ChatGPT, etc but I found these tangential points interesting:

> Although Python and Java continue to be the most frequently used programming languages on the O'Reilly platform, the report discovered a substantial growth in interest in Go and Rust, with both experiencing 20% and 22% growth

> Interest in coding practices experienced 35% growth, the largest jump among software development topics, indicating that developers are highly motivated to improve their programming skills. In the same topic group, quantum computing showed a 24% year-over-year gain.

> Developers frequently looked to grow their “soft skills,” accessing topics around project management (47% growth), professional development (37% growth), and communication (26% growth).

Did you try it out?

Aren't you impressed how well it understands your questions?

I have never seen a tech demo as good as chatgpt.

I don’t know about GP. But my issue with ChatGpt is that it always has an answer. Even if the answer is factually incorrect. It’s impressive no doubt. But this notion that there is some intelligence behind it is just baffling to me.

I think chatgpt will cause a LOT more problems than it will solve.

One problem is that the average person thinks it is good. From a technical view I don't see anything interesting, it is garbage in garbage out. It depends on the quality of the training data.

I've used it quite a bit while developing and it regularly gives very wrong answers. Sometimes giving me java code and then getting the library dependency wrong. Or giving a version for the dependency that doesn't even exist.

I've asked it questions about Demon's Souls while playing and it told me quite confidently that the NPC the Filthy Woman does not exist (she does).

It feels more like a parlor trick for children and it's forced me to waste extra time verifying what it tells me using secondary sources.

It sounds like you're trying to use it like Wikipedia when it's really more like Microsoft Word.

Asking it for plain literal facts is the weakest use of it IMO.

Why chatgpt and not GitHub copilot?

I knew that this was gonna be Hot Bot Summer.

Not to be mistaken for HotBot Summer back in '96.

'Hot AI summer' certainly has a dystopian ring to it.

I'm sure this is in no way a reference to the Terminator apocalypse. That also had a 'Hot AI summer'.

The world has never been closer to being glassed by the supernuclear powers, 2 out of 3. China claims to be rational. Video of the pretorian guards escorting the qua Anon Shaman Jacob Chansley through the Government building reinforces learning on the weighted graphs or artificial neural nets.

I still think there is a high probability the whole bubble will burst explosively by the end up the summer. We've seen an unprecedented self-reinforcing wave of hype and commensurately inflated expectations. When it falls apart, it's going to fall hard. I'm not saying there's nothing to the tech, just that expectation have become so inflated that whatever new developments actually happen will never he able to keep up. I'm still betting on AI winter (with a long fall driven by inertia in governments and big institutions that didn't get the memo). This is going to be a highly exaggerated version of what we've seen with blockchain (edit: with the qualifier that while blockchain will go to zero unless someone finds a use for it, ML will settle to its nonzero real utility)

Under 6 months seems way too small for a tech hype cycle. This AI stuff has more staying power than Crypto and that has had a 10 years or so of cycles so far. Amazon, Google, Microsoft and Meta will circle the wagons around AI for years, consulting firms will start offering integrations to small players, there will be many startups. We are too early on this to be predicting AI demise.

I'm following ml papers, ml work etc for a while.

The progress is crazy.

The use cases popping up are already here.

Not betting on ai/ml right now is something someone can't afford.

And I'm not easily be swayed.

But alone the fact how good Charcpt understood what I wanted it to do and seamless switching languages is crazy good.

And ml is everywhere: voice synthesis, OCR, image analysis, image synthesis, text understanding, driving, ...

This is the new way.

> And ml is everywhere: voice synthesis, OCR, image analysis, image synthesis, text understanding, driving,

Just to be clear, in order those are: RNNs/CNNs, CNNs, CNNs, not something neural nets do, and multiple AI systems with widely different architecture.

The progress in GPT has very few aspects that pass over to those.

Reply on news.ycombinator.com