The (Open)AI Bubble

Open topics
1

Thought this post was interesting. It captures how OpenAI specifically has been making some wrong bets, and has actually stopped providing as much value as perceived as a company.

I’d say that AI is here to stay, but in terms of competition, OpenAI has a lot to overcome.

1 Like
2

So this is wildly interesting to me. Essentially its saying “OpenAI partnership = risky, expensive burden.”

I’d argue this is highlighting a skeptical market…. which is definitely something to keep our eyes on. :eyes:

1 Like
3

I wonder if it actually reflects Sam Altman’s personality :joy:

4

The AI industry is definitely inflated and OpenAi has led the charge for the same. They have built something that is incredibly expensive (in terms of time, cost and resources) and is not getting the results we had all hoped for. That said, to get the real result we need to go through a process of reimagination for all our workflows. There is too much to write about and too many thoughts in my mind with regards to the value we get from AI. When I get some time I will try to put them together and ask for comments.

2 Likes
5

This would be really awesome to see @nikhil_mahen. A lot of thoughts are a good thing, haha

6

@nikhil_mahen I am sitting on the edge of my seat for this! I find this area to be super interesting now that we’re seeing more and more investments ….. (& building data centers all over) and still skepticism is on the rise.

I think there is value, surely - so I would love to read your thoughts on this!

1 Like
7

So here are a few issues with the normal AI value creation from the top of my head :

  1. OpenAI/ larger Gen AI bet 1: there is enough content in the internet to train specialised llms. Turns out its not true when you are creating an LLM with high temp and a monolithic vector approach. Some Chinese startups have shown the MoE approach or a specialized SLM approach sitting on top of LLMs to give better results but it undermines OpenAI a bit.
  2. The value generated should have been straight forward from just prompting. By their design LLMs are limited with the value they generate. However they arent. Few specialist teams have the ability to implement a data pipeline to mental gymnastics to get the best quality output.
  3. Open AI are betting on the world building solutions on top of their platforms to get the most value: in the long term this is a good strategy for open source llms but is costly when one considers Open AI APIs. Moreover as good solution architects for this are few and far, the value you get from APIs isnt enough.
  4. Over priced and overvalued: Open AI and other AI leaders have fallen in this economic trap which i am sure is a syndrome. Their offerings are too expensive (as it stands today) to justify all costs and values limiting experimentation. However, the actual costs of running these are much greater in terms of manpower for training, infra and electric resources. So they are too expensive to run and cost too much to create large scale value. Which is why most people try to derive as much value from the free tier.
  5. Gen AI hasn’t penetrated actual workflows till the extent one would have hoped. Successful agentic workflows in enterprise are few and far. To keep up with the media jazz, open ai has prioritized building B2C solutions over B2B. I believe that to be a mistake. B2B give you more scale for the same usecase with a singular client to get coherent feedback from. B2C usecases are IMO of much lower impact and not valuable enough.

A lot more from the value generation perspective but I will stop for now.

3 Likes
8

@nikhil_mahen This is an awesome, well written breakdown though. It aligns to my conceptions around the technology as well.

I actually think that the magic is happening outside of OpenAI. These smaller, cheaper models that have been specialized, like Cursor’s Composer, or even a funny one, The Big Pickle by Dax, have reached a point of “good enough” to do the 60-70% of the programming work while costing almost nothing (basically they’re achieving 80% success rates of larger models like Opus and GPT-5 all while being faster).

1 Like
9

@Kevin_Schumacher tagging you! I know you’ve been tracking the perceived value of AI in the workforce - thought Nikhil’s blurb here might be interesting + the context of Ben’s article.

This is a killer breakdown here, thank you @nikhil_mahen - the one that sticks out the most to me is forsure “Over priced and overvalued” - will be interesting to see how this shakes out.

2 Likes
10

“Successful agentic workflows in enterprise are few and far. To keep up with the media jazz, open ai has prioritized building B2C solutions over B2B.”

One reason may be that there is a perceived limit or ceiling for the quality of the AI output right now, and for a consumer market that ceiling is a lot easier to accept because users don’t have as high expectations. Quality and accuracy is more necessary for B2B users, so the AI builders have to break through that ceiling, which is not easy work (speaking from our own experience building our new AI tool :sweat_smile: ).

3 Likes
11

Thought this podcast was interesting. I think most of these companies are banking on scaling being the solution to AGI.

This hackernews post has a few gems that are closer aligned to where I think we’ll be in the next few years

12

Great quote from the hacker news thread:

"As someone who’s been integrating “AI” and algorithms into people’s workflows for twenty years, the answer is actually simple. It takes time to figure out how exactly to use these tools, and integrate them into existing tooling and workflows.

Even if the models don’t get any smarter, just give it a few more years and we’ll see a strong impact. We’re just starting to figure things out."

This seems about right. Working teams that know how to make great products like this are limited, but the knowledge will continue to create better products.

1 Like
13

Hacker news typically has a a lot of gems and some pretty smart people in there.

I fully agree with this take, and we definitely have to be in front of the wave (yes, the wave is still coming, even if it’s smaller but in longer duration).

14

This is an interesting signal. Agents are definitely the way forward, but getting companies to move forward with a workflow solution rather than just slapping a ‘generalist’ AI feature on will be a transition.

1 Like
15

Yep. The problem is that everyone (~95%) sucks at integrating AI into existing products.

I think that we’ll see new trends appear Over the next couple of years