How AI Got a Reality Check

Two years ago, OpenAI’s ChatGPT became the tech industry’s biggest product in years. Now, leading developers like OpenAI, Google and Anthropic are finding their models aren’t improving as dramatically as they once did and profitability remains elusive.

00:00 Introduction
01:28 Origin of the AI Boom
02:39 Limits of large language models
04:38 Synthetic data
05:10 Cost pressures
07:22 What’s next?

#AI #technology #business
--------

Bloomberg Originals offers bold takes for curious minds on today’s biggest topics. Hosted by experts covering stories you haven’t seen and viewpoints you haven’t heard, you’ll discover cinematic, data-led shows that investigate the intersection of business and culture. Exploring every angle of climate change, technology, finance, sports and beyond, Bloomberg Originals is business as you’ve never seen it.

Subscribe for business news, but not as you've known it: exclusive interviews, fascinating profiles, data-driven analysis, and the latest in tech innovation from around the world.

Visit our partner channel Bloomberg Quicktake for global news and insight in an instant.

[sleepykitten]

This dropping 8 hours before o3 is funny.

[たなぬにあさか]

Who are these reporters talking’s like experts in the field?
Really hate this format, it’s lazy. Just interview the real people in the topic.

[Vader-twgg]

"The moves I did were calculated,
but man am I bad at math."
-AI (probably)

[ObeAroundTheGlobe]

chatGPT wasn’t the AI breakthrough - it was the transformer architecture

[ydinuda]

Just a FYI. the openAI reasoning model (o1), doesn't give better structured output because we give them more time to "linger" on the problem. But because it re prompts itself. to have multiple tasks, and execute based on said tasks. It's a technique called "chain of thought" and that concept has been there ever since we had llms.

[Patapom]

5:00 Feeding LLM-generated bullsh*t for training other LLM's... What could go wrong?

[connorutz]

It took 8 hours for this video to be outdated

[BigK-]

Maybe should've waited till OpenAI's 12 days were over.

[marekkowalski]

Promising technology got over hyped, only to turn out it will develop notmally over a decade? No way.
There is a quote I like: People tend to overestimate how much we can advance in a year and underestimate how much we can advance in ten years.

[averytucker]

For me, my biggest problem with these AI tools is the fact that they don't really have much practical use in an actual creative work flow. Every time you see these AI tools being used. Its more novelty type things then workflow. I don't see anybody using it in a way that feels like something you'd actually be using at your job at all.

Like as a digital artist and a photographer. There's literally nothing AI related that I use in my workflow. That isn't AI that has been established decades ago. All this new ai from the 2020s is pretty useless.

[matt.stevick]

i am starting to see a pattern that bloomberg is so clearly against ai.

[milliamp]

As someone pretty close to this space this is pretty accurate but related to 5:20 about models costing more to train I should point it pushing the limits of state of the art is expensive but catching up to the status quo is actually becoming much cheaper as the SOTA begins to hit a plateau. Meta 3.3 70B beats GPT-4 and GPT-4o for instance and it was a fraction of the cost to train. Similar results can be achieved with smaller, faster, and less expensive models today. Outside of that point most of the video was generally correct.

[vroomik]

5:45 "It hard to say if anyone's making money"...Maybe some small startups do, but if OpenAI, Microsoft or Google would be making money, they would be loudly posting their profits. They don't.

[Bamboo_beach]

Quantum computing and the new Willow chip will change all that fast

[zrblank]

Shout out to the comment section. I scrolled down hoping someone would call Bloomberg out, comments didn't disappoint!

[iinarrab]

Instead of interviewing people who works in AI, idk who did they interview.

Whats hitting the wall are human expectations — both the (dumb) investors and (hyped) humans.

However, the research has always been moving fast. As someone who has been doing AI since models were able to generate unrealistic human faces until the first breakthrough of StyleGAN, AI has been rapidly moving in the research department.

[octavioavila]

6:40 I appreciate the nuance she brings to the conversation. Her analyses were captivating and her way of speaking communicates intelligence.

[gptBreeze_io]

01:34 AI’s Historical Context and Recent Breakthroughs
03:25 Top companies like OpenAI, Anthropic, and Google are facing difficulties in improving model performance. High-quality, curated data is increasingly scarce, making it harder to further improve AI capabilities without relying on expert-level data and specialized training.
04:25 Synthetic data is uncertainty about its reliability compared to human-created data.
05:03 AI development is expensive: Training a new model can cost 100 million, potentially rising to100 billion.
07:04 The potential for Artificial General Intelligence (AGI) raises concerns: AGI could surpass human intelligence, leading to uncertainties about the future role of humans in an AI-driven world.

Summary by GPT Breeze

[Gepprocurement]

The rapid rise of AI, as illustrated in the video, underscores both its transformative potential and challenges. From the meteoric launch of ChatGPT to the growing concerns about cost, scalability, and data sourcing, AI’s evolution (and adoption) has been cautious and nuanced.

For organizations exploring the adoption of AI, ensuring strategic alignment and ethical implementation is key to unlocking its full potential.

[borisadimov]

How Bloomberg got a reality check in 8hrs