Is AI Slowing Down?
17:01 Share

Today, on the A I daily brief is A, I actually possibly slowing down before that. In the headline sales force hires one thousand sales people to sell A, I sales people. The AI daily brief is a daily podcast and video about the most important news and discussions in A I. To join the conversation, follow the discord link in our show notes.

Welcome back to the AI daily brief headlines edition, all the daily AI news you need in around five minutes. Today we are going to rip through a set of stories. As the main episode is a little bit longer than usual.

We kick IT off for the story from salesforce. Salesforce is pushing the agent story more than any other company out there. You might have heard CEO mark any off absolutely, positively screaming about the fact that the assistant era of A I has been a big let down because, of course, it's really and was always really about agents all along.

The news today is that sales forces hiring one thousand people to push their new AI agent platform bending offset that the hiring surge aimed at making the most of the amazing momentum of their new product adding agent force became available just two weeks ago, and we're already hearing incredible feedback from our customers. One of the important pieces of context is that sales force has been downsizing their sales team over the past two years. So there is clearly a lot of excitement around this new product line.

Of course, the joke everyone is making on twitter is some variation of what Thomas smaller wrote, sales force higher one thousand sales people to sell their a sales person. Just writing that out sounds like a joke. But still, I think it's a really positive sign that there is so much demand that this is the actual direction they're headed.

Meanwhile, over an elon world, x is testing a free version of X A S rock chapt. Until now, the AI features of x only available to paying customers. However, over the weekend, IT seems that rock was available to free users in select markets.

Tech runs confirmed that all of new zealand had access to grow, and IT seems that australia also has access. Now usage is limited with thirty text queries every two hours and three image analysis questions per day. But there's still a lot of excitement.

Doctor warfare rights if gog becomes free, I will delete ChatGPT over in the land of google. According to the wall street journal google news, exact salleh per cash has resigned. The resignation comes as news publishers rethink their relationship with big technical A I era.

Publishers have ve seen traffic and revenues decline, with many blaming A I E for serving content summaries rather than driving clicks. The issue has become even more contentious with google rolling out their AI overviews globally for cash was originally brought in two years ago from the washington post to bridge the gap between the two industries and therefore many. You are wondering if his departure is a signal that the relationship is showering even further.

Then again, there are clearly structural changes underway at google following a year mark with losses and enter trust lawsuits. So it's hard to know exactly what's going on. One other interesting story out of google the companies is deep mind division has unexpectedly released alphago three.

The source code model ates for the protein folding A I are now available for academic use, which could further accelerate scientific progress and drug development. The surprise announcement comes weeks after the systems creators demo service and john jumper were rewarded the nobel prize in chemistry for their ground breaking model. Alpa l three seems like another paradise shift for biological science.

Version two could accurate predict protein structure, being said, nothing but there. Mino acid sequence. The new version can also model the complex interactions between proteins, D N A R N A small molecules.

This should massively expand its useful ness and drug discovery and disease treatment. Traditional methods of studying these interactions often take months of lab ork with no guarantee of success. This Marks an important shift in computational biology.

A I methods now outperform our best physics space models and understanding how molecules interact. Essentially, alcohol has gone from a specialized tool to a more general sis army ife for researching molecular biology. Over in the market realm, U.

S. Companies are set to spend thirty billion dollars on data centers as A I drives a construction boom. That amount has more than doubled since twenty twenty two in ChatGPT was first released.

Data center investment is now bigger than every other major infrastructure category, including hospital, schools, hotels and transportation. A smaller category than each of these until late last year. Investment manager K K R believes the trend is expensive.

Al expecting IT to hit two and fifty billion per year globally within the next three to four years, although stick around for our main story, as there could be some implications on that front. That, however, is going to do IT for today's headlines. Next up, the main episode today's po de has brought you by plum, want to use A I to automate your work to do plate AI workplace by simply describing you.

No coding or A P is required typing out I analyzed my zoo meetings and send me your insights in notion and watching IT come to life before your eyes, whether your an Operations leader, cod sonat, three point five assembly A I and many more, don't the technology hold you back? Check out, use plum. That's plm with a bee for early access to the future of workload donation.

Today's episode is brought you by van tab. Whether you're starting or scaling your company security program, demonstrating top notch security practices and establishing trust is more important than ever. Penta automates compliance for I S O twenty seven, O O one soc two gdpr and leading A I frameworks like I S O forty two thousand one and N I S T A I risk management framework, saving you time and money while helping you build customer trust, plus you consume line security reviews by automating questionnaire s and demonstrating your security posture with a customer facing trust center.

All power by vent to A I over eight thousand global companies like LangChain l AI in factory A I use vented to demonstrate A I trust, improve security in real time, learn more eventide c com flash N L W that's ventadour com slash N W today's epo de is brought you by super intelligent. Every single business workflow and function is being remade and reimagine with artificial intelligence. There is a huge chAllenge, however, of going from the potential of AI to actually capturing the value.

And that gap is what super intelligence and is dedicated to filling super R N. Li ligon accelerates AI adoption and engagement to help teams actually use A I to increase productivity and drive business value. An interactive A I use case registry gives your company full visibility into how people are using artificial intelligence right now, pare that with capabilities, building content in the form of fu tori's learning paths and a use case library.

And super intelligence helps people inside your company show how they're getting value out of ai while providing resources for people to put that inspiration into action. The next three teams that sign up with one hundred or more seats are going to get free embedded consulting. That's a process by which our super intelligence team sits with your the specific use cases that matter most to you and helps actually ensure support for adoption of those use cases to drive real value.

Go to be super di I to learn more about this. A I enable ment network. And now back to the show. Welcome back to the AI daily brief.

Today, we are having a really fascinating conversation that has implications for everything in A I, from safe to the business, of IT, to technology frontiers, to applicability, to regular life, and all sorts of interesting things bottled up in this one conversation around whether th Epace o f d evelopment o f A I i s a ctually, now this is not a new conversation, is something that has been debated for a very long time. And by the way, there are wildly divergent opinions here. There are some folks who think that you can just add more compute and more data and get more performance models.

There are others who think that there is no path to A G. I. With the current architecture we're using. This debate is why people like Young the kon can with a straight face, say that they believe that A I is dummer than a cat. And of course, the implications of that view are significant.

In a particular case of the cons point, IT means that, to him, all of these existential export type concerns are way, way overblown. Now, what no one is arguing is that, A, I isn't extremely valuable even in the state right now. One of the things that i've often said is that even if A I stopped developing in this moment, I would still take years for the world to adopt and adapt to the new processes that have been made available by just the technology that we in this moment.

However, IT still would be fairly significant if, in fact, the of the I improvement was decreasing significantly. Are right to let's start by discussing where this conversation, or at least this iteration of the conversation actually came from the till the R, S, that the information reported that OpenAI are apparently looking to develop new strategies to deal with the slowdown in A I improvement. Back in may, OpenAI C E O C A M.

Altman had told staff that he expected their latest frontier model, which they're calling a yan internally nt GPT five. He believed to be significantly Better than last year's flagship model at the time, OpenAI had only completed twenty percent of the training process for iran, and IT was apparently already on par with GPT four level. However, now that there have been a few more months of work, things are starting to look a little bit different.

According to the formation sources, employees who have tested the on model found that although the performance does exceed the current models, the jump isn't nearly as profound as the improvement from, for example, GPT three to GPT four. There are even some use cases where the model might not be consistency Better than the previous state of the art according to one employee. While ryan does Better at language tasks, that doesn't necessarily outperform when IT comes to coding, which, given how prominent coating is as a use case, could be a significant problem.

Now what this is leading to is a different approach to thinking about scaling the information rights. In response to the recent chAllenge to training base scaling laws posed by slowing GPT improvements, the industry appears to be shifting its efforts to improving models after their initial training, potentially yelling a different type of scaling law. Another approach OpenAI has apparently created a foundations team to figure out how the company can improve their models given what is increasingly a scare supply of novel training data.

Those strategies reportedly include training or yon on synthetic data produce by other AI models as well as doing more to improve the reasoning process during the post training process. One of the big chAllenges is that at this point, the current frontier models have effectively been trained on pretty all of the internet data. Still, by far, the more interesting shift here is about thinking about scaling simply as a matter of free training, to instead thinking about the post training implications.

And this gets back to something that we've talked about a lot here, the idea that, oh, one which is, of course, OpenAIce mo re ad vanced re asoning mo del. Is fork on the evolutionary tree of alms. Sam altman, in when he announced IT, was clear that this reasoning model represented a separate path from GPT4。 At the time, I was assumed that open a eye would parallel process these paths.

But increasingly, IT seems like they're leaning into the reasoning path. And by the way, it's not just open a eye who's dealing with this problem. A few weeks ago, alex heath of the verge reported, i've heard that the latest gemini model isn't enjoying the performance gains that google had hoped for, though I would still expect some interesting new capabilities.

The chatter i'm hearing an AI circles is that this trend is happening across companies developing leading large model. Former open a eye chief scientist daily a sits caver spoke to this issue, stating the twenty tens were the age of scaling. Now we're back in the age of wonder and discovery.

Once again, everyone is looking for the next thing. Scaling the right thing matters now more than ever. And this once again gets us to what cheering award when your Young, the colon was pointing out in last month.

Street journal peace, the W S. J wrote the on thinks that the problem with today's A I system is how they are designed at their scale. No matter how many G, P, U. Tech giants can to data centres around the world, today's eyes are, aren't going to get us to artificial general intelligence. There are also questions around whether synthetic data will actually create their own inside of new problems, although that far from clear as well.

But overall, IT still appears that open an eye is planning to double down on their reasoning pathway last month that an appearance at tedy eye sentence is go OpenAI, researcher nme Brown said. IT turned out that having a about thing for just twenty seconds in a hand of poker got the same boosting as scaling up the model by one thousand x and training IT for one hundred thousand times longer. And while this does open up a new pathway for improvement and also implies having a lot more computing power to service answer, ital said the shift will move us from a world of massive pretrail clusters towards inference clouds, which are distributed cloud based services for inference.

Video C E O jen hung also referred to this pathway, stating, last month, we've now discovered a second scaling law, and this is the scaling law at a time of inference. So where has the discussion been? First of all, there has been some confirmation from others that this is bubble behind the scenes on paleo route.

Heard a league from one of the frontier labs, not OpenAI. They reached an unexpected huge world of diminishing returns, trying to brute force Better result by training longer and using more and more data than what is publish publicly for others who have been saying the stuff for a while, they use this as a moment to reinforce their point. Professor patr to mingo wrote, scaling laws are s curves not expended tiles? inc.

Professor is in. Davis explained that a little further. He said this maybe the most important strategic question about A I is the rate of improvement available and starting to decline. Are we seeing maturation in the s curve? The report from the information gives us some data, but I think we really don't know.

And is that odds with many insiders working on foundation models, especially since the affected performance of election systems will depend on complimentary that extends usefulness and perhaps many years of individuals and organza finding the right use cases and configuring their workplace to leverage them. Others, including the information, found a Jessica lesson. Wonder what the implications are for the data center business, SHE wrote.

If this continues, the slowing rate of gains and training elements has huge implications. One thought I had this morning, while being at the bleeding agent, ships matter as much over time. Will china therefore catch up to the U.

S. More quickly? Others put IT out that, of course, there was always going to need to be something new.

Sigh on twitter rights pre transformation architecture, I had hit a brink wall. More insects in architectural tween will be needed to break successive barriers for improvement. That would be unwise to imagine that current alm is the final solution.

Perhaps the most salient message though, is just that we need to think about scaling in training in a slightly different way. Australia entel on twitter, flash x says, A, I hitting a road block, give me a break. As if the diffusion transformer breakthrough hadn't happened, as if a one reasoning is in ground breaking on its own.

The only thing reaching a plateau is retraining. And that's a milestone, not a ceiling Price. Training was never meant to be the holy grail of neural network to mize.

Ation will inevitably move towards models that can actively learn for a reason hater at slow developer rights. Even if current alams can scale to A G I through computer increases alone, it's not a problem. OpenAI a one shows further innovation are possible and we haven't reached inference compute yet.

Things are progressing even if fels don't magically scale to A G. I. With more compute professor Shelley palmer rights. There's been a lot of chatter about the end of alms or alam starting to fail. Those are only headlines, click bait, really.

If you dig a bit deeper, you will read that there are several schools of thought regarding how to efficiently scale the foundation models. If the goal is A G I, then just add in computer wer to pretrail ing may not be the best path to follow. Instead, researchers are expLoring an alternative called inference scaling to achieve smart a influence.

The process when an A I model generator outputs and answers can be optimize zed by having models, quote, quote, think through multiple possibilities before settling on a response. This approach ables complex reasoning during real time use without increasing model size. Open the eyes recently launched a one model as a good example.

By enhancing inference, a one can tackle tasks that demand layer decision making, such as coding, a problem solving in ways similar to human thought test time. Computer techniques make this possible. Allowing models to dedicate more processing to chAllenging inquiry is needed.

A move towards inference focus distributed cloud base services instead of large centralized training clusters might create a more competitive chip let. While in video is the go to chipmaker for pre training hardware, there are a bunch of chipmakers AMD, intel, etta. They make hardwork suitable for this new method.

Inference scaling the key takeaway is simple. L EMS are not fAiling. They are evolving. Sensationalist headlines aside, this is how product development works. Every dan shipper also pointed out that the framing of all of this was very problematic and that I didn't really reflect the conversations that he's having with people inside the labs.

He writes the message that this headline conveys is at odds with what people inside the big labs are actually feeling and saying it's technically correct, but the take away for the casual reader that AI progresses is slow is exactly the opposite of what i'm hearing. And indeed, ultimately, there was so much chatter to that effect that the information actually ended up clarifying a little bit. Stephane paleo, a rights seeing lots of discussion after our scaling laws article this weekend.

To be clear, we're not saying the world is ending, just that researchers are having to find new ways to improve alams I E test time compute that LED to another article, goodbye GPT, hello reasoning. Oh, and this basically gets at the point that this isn't about a halt. It's about an alternative direction and a new path that shows more promise than the previous.

Still, it's a super interesting conversation, one that we will continue to follow closely. For now though, that is the story. Appreciate you listening as always, and until next time, peace.