Hacker News Reader

Comment by ChicagoDave | Hacker News Reader

This is only the beginning of AI-centric offerings that were oversold and will be delayed or quietly abandoned.

LLMs are nice for simple things, but they’ve already reached their limits. No amount of data will solve the iteration and complexity problems.

View on HN

Every month I am in meetings where LLMs are being considered for applications they are absolutely not the right fit, but the answer to my concerns is "we need more AI advocates". These conversations are led by people who never actually read a single paper on LLMs or tried them in real life. They have no idea about risks, but plough on because their clueless bosses told them to come up with a plan to use AI.

I remain very skeptical that most companies or products can or should integrate with LLMs - and I say this as someone who works on a LLM-based product!

Overall I feel like our industry has lost the plot to a large degree. Hype has always to some degree exceeded the merits of the technology-of-the-month, but the last few cycles have been truly extraordinary in terms of the gap between the breathless hype and the reality of the tech. It's LLMs now but before it was crypto.

It just seems like we're stagnating as an industry, and rather focus our efforts on the hard R&D needed to reach the next Big Thing, we've decided it's much easier just to focus on cults of personality combined with vast over-hypedness.

Investors want their 10x returns. It's worse than you expect. They know nothing and reject scientific papers that discuss problems with LLMs. They were told by marketers that all problems with LLMs can be solved by feeding them good data... which is a sneaky way for LLM operators to obtain access to data they should never be allowed anywhere near. People with the collective brain of a Handforth Parish Council are working on strategies for using AI on PII data. Imagine the cretins from the Office discussing AI and you get the picture.

Honestly, people just need to touch a hot pan every now and then. Let them slap their LLM onto something low-stakes and experience the results for themselves.

Everyone does, it's just that some of us have the decency to do it in a homelab or on a cheap proof of concept first.

Nobody has argued that this feature doesn't work. In fact, it probably works really well which is why Microsoft has been pushing it so hard.

Wait how do we know that they’ve reached their limits?

If you understand how vector databases work, you realize that LLMs can only tell you what is in the data. There’s no such thing as perfect data except maybe for things like tax preparation or tic-tac-toe.

Inherently, no LLM can provide a solution based on a creative logic it does not possess.

It’s that creative logic that has not been invented/automated yet.

The delay of Recall has absolutely nothing to do with technical limitations.

The marketing was 100% trying to sell AI built into a new version of Windows. They completely jumped the shark because of technology.

That they weren’t thinking it through is endemic of everything going on relating to LLMs.

Reply on news.ycombinator.com