AI Hype is completely out of control - especially since ChatGPT-4o

preview_player
Показать описание
Now that ChatGPT4o is out, it's time to revisit my previous thoughts on AI, and discuss what might be happening in the future, and what evidence we have of what might happen.

00:00 Current brief status of AI
03:21 ChatGPT-4o is faster. at least
03:30 The Multi-modal interface isn't new
04:00 What matters for disruption?
04:48 Science and measurement
05:46 What counts as evidence?
07:54 ChatGPT-4o benchmarks
10:39 Psychology of Humans and AI
11:28 Dark Patterns
12:35 History of AI demos
18:40 New article about AI Hype
19:54 Conclusion

Sources from Video have been moved to a pinned comment, since the description field ran out of characters.

Рекомендации по теме
Комментарии
Автор

Moving Links and Sources to Here since the Description ran out of characters.

# My last video about how I expect AI to impact the Software Industry over the next few years

# BP is saying they're now needing 70% fewer coders

# Bloomberg on the number of times "AI" was mentioned on earnings calls

# Job impact of the 2000 .com bubble

# instructions on adding voice to chatgpt from Feb 2023

# Explanation of how multimodal embedding is "used in the exact same manner as its textual counterpart"

# 4o vs 4Turbo benchmarks:

# Screenshot of ChatGPT-4o telling me Stave Jobs didn't address Radio while Perplexity took me right to his 1983 quote

# Andrej Karpathy on the SEAL benchmarks

# How is ChatGPT's behavior changing over time?

# ChatGPT5 as a whale

# humans are predisposed to be particularly gullible about the sentience of AI

# papers on the long, long history of humans believing that things are sentient when they aren't
# from weather to pet rocks

# The "Eliza effect" has been known since the 1970s

# "Dark Patterns" in UX design

# Dark Patterns in LLM ChatBots

# synthetic voices and "Cuteness" as Dark Patterns

# Tesla faked a self-driving demo in 2016

# Independent recreated a Tesla demo

# Google Duplex faked demo

# Google Gemini Faked demo

# Researchers refute Google's claim of DeepMind creating 2.2M new materials

# Google Overview AI debacle

# Amazon Fresh's "Just Walk Out" was 1000s of remote workers

# Amazon's "Mechanical Turk"

# GM Cruise's 1.5 remote operators per "autonomous" vehicle on the road

# Facebook's "M" Chatbot used humans to answer 70% of user questions

# SEC charging companies with "AI Washing"

# AI Claiming "transparency" in AI, then a whistleblower detailing how they allowed deepfake porn of Taylor Swift

# Rabbit R1 is just an Android App (and doesn't do what it the demo said it could):

# Humane AI Pin:

# ChatGPT-4 was claimed to be better than 90% of people who took the Bar Exam, but was really only 15%.

# "Turns out the viral 'Air Head' Sora video wasn't purely the work of AI we were led to believe"

# "Remember the ballon head Sora video — it wasn’t all AI generated after all"

# "Uncovering The Reality Of The Sora Film Hit: Allegations Of Deceptive Special Effects Manipulating The Audience"

# Devin's company lying about its Upwork job

# Interview with former OpenAI board member about the "Toxic Culture of Lying" at OpenAI

# Story in WIRED claiming "But the Demos Aren't Lying"

InternetOfBugs
Автор

The market correction is going to be brutal when the MBAs realize computers aren't magic

xevious
Автор

Programmer: Pretend to be alive
LLM: I am alive
Programmer: What have I done

buff.berserker
Автор

Wow, it's pretty shocking that an industry that relies almost entirely on venture capital would be motivated to lie. Maybe, just maybe, there's more to being sentient than doing simple math problems insanely fast.

Jeremyak
Автор

Elizabeth Holmes must be kicking herself that she got in the wrong racket, when this goldmine was just a few years ahead

jamesarthurkimbell
Автор

companies using AI to argue about "we need less devs bc of AI" is absolute bullcrap. That simply shows that CEOs of BP and alike keep people as idiots. Whatever disruption chatgpt brings (if there is) it is in infancy at best

This hype will soon plop in a fascinating and spectacular way.

New AI tools will certainly find their niche, just like it happened countless times before in history. Nothing more

snowbarsyk
Автор

I'm just worried that companies will use AI as excuse to pay devs less and less.

mnchabel
Автор

I think a lot of why people are more tempted to attribute thoughts and feelings to AI is because we call it "Artificial Intelligence" which sets expectations of some sort of higher level sentience like in the movies. Machine learning would be more accurate but I guess less marketable.

AkilanNarayanaswamy
Автор

A big problem is any GAI application is that it becomes useless for any kind of objective work when it hallucinates. Even if only 1% of responses are random nonsense, that makes it basically impossible to use reliably without a human handler watching over it constantly.

BlazeMakesGames
Автор

Sir, why are you out here telling facts!? I refuse to listen to facts! Just hit me with some sweet sounds of Elon Musk's hyperloop, Mars' colonies, robotaxi, flying cars, and give me some Elizabeth Holmes... whisper sweet nothings into my ear and take me money

QuantumEX
Автор

I'm a freelance academic editor and AI has had a major negative impact on the sector. I expect a bounce-back because unless they are experts, folks writing in another language don't actually know what the AI translations mean. Because AI isn't self-correcting, the punters only find out when their paper is returned with a multitude of abstruse journal queries that the bots pretend to understand but can't fix.

BeXEllenttoeachother
Автор

And this is why I've been calling the current trend of AI "Active Incompetence."

BrunodeSouzaLino
Автор

Thanks for the video, and just wanted to share my story and opinion.
I'm just a casual backend developer, very occasionally using AI, but a few weeks ago I started my project and tried to use ChatGPT-4o as a frontend React developer, as I have neither experience nor desire to deal with that. It took me more than a week to deal with one page, though my prompts were thorough and vast, and I tried uploading files and sending screenshots of what he has done and what I try to get from it AND EVEN THEN I had to debug and fix stuff. And still I'm afraid that if I show this code to an actual frontend-dev, that would be a shocking experience. To conclude, I want to say, that even that I'm not afraid of AI getting our jobs'n'stuff in the nearest future, all this hysteria makes a rather unpleasant background noise.

MatreshkaVodka
Автор

I did a web scraping test last week using Jina and ChatGPT4o... Jina scraped about 200kb of text from a hideously complex quarterly report. I fed the data to gpt4o and asked it to create a json object containing all the people mentioned, location names, company names etc. as lists in the json object, also to include the paragraph in which they appeared. The answer was accurate and instant. Ollama3, CodeLlama and another I can't remember all failed to even produce a valid json object. For me AI generally sucks, but for this kind of automation gpt4o at least produced an excellent result... and Jina doing the scraping was spot on.

m
Автор

Thank you for mentioning the "ELIZA effect" and the non-tech industry stuff. I'm originally a philosopher of computing working in cyber security, and getting people to think about the cognitive psychology, literary criticism, philosophy, neuroscience, etc. that is related is incredibly hard. Philosopher Daniel Dennett has a piece in _The Atlantic_ calling upon the world to ban counterfeit people - and this from a scholar who spent his career defending AI from "it is impossible" critics, etc. I have been thinking one way to phrase the security control pattern we need is "deintentionalization", which is, as our host says, the exact opposite of the direction the hype is going.

logiciananimal
Автор

We are in a classic hype cycle, this is how tech investment works. The only way to get the funding to take a shot that has 99% chance of failing is to sell this story thats so big and bold that nobody can look away. The most recent hype cycle which was very similar in nature was crypto. Exactly the same level of madness which included 3 things - popular culture, paradigm shift, scarce next gen hardware. Its exactly the same with AI.

However I am not seeing revenues jump for companies releasing AI features, nobody is cleaning up the market or taking away market share from competitors. What I do see is an explosion in small startups that are offering similar functionality as big tech SaaS. Ultimately all the margin is going to be competed away. So this market will be owned by the lowest cost labor market.

bikesbeersbeats
Автор

Remember when that one "whistleblower" in 2022, a google engineer apparently, was convinced that google's chatbot was sentient? I don't think it was explicitly said, but I'm pretty sure we can say with some certainty that that was google gemini. And yeah, no, it most certainly is not.

tetravisum
Автор

As someone who has developed around LLM capabilities, I’ve come to think that LLMs are natural language processors on steroids — because of the attention mechanism. Attention — that allows to capture meaning — together with multimodality, really plays with the human brain. Having said that, it is a great performance improvement tool, with some clear use cases — I’m thinking of RAG. And there are others in the pipeline. Imagine RAGs coupled with reinforced learning.
Are LLMs great? Yeah. Is this intelligence? Nah. Is this current technological iteration reaching a plateau? Absolutely.

flcor
Автор

"cuteness is a dark pattern"
Well, Japanese products have always been adopting this since the booming of anime 🤣🤣

zookaroo
Автор

I'm noticing some similarities between how AI and games are marketed. Game studios want to give the best possible impression of the games, and in the earlier days they tended to just provide prerendered footage that was nothing like the game itself. Gamers get fed up and just want to see actual gameplay footage rather than prerendered trailers, the studios somewhat listened, but you should be vary of even gameplay footage as it's cherry picked.

I'm specifically reminded of when the console wars was at an all time high and the hype was around the upcoming ps3 and xbox360 was intense, sony decided to reveal gameplay footage of Killzone 2 at E3 2005 which blew peoples mind with how ahead of its time it looked. But in the end it turned out it was all prerendered.

But if you compare the fake trailer to what real games look like today, we've come a long way. :)

CapsAdmin