OpenAI COO thinks AI for business is overhyped | Hacker News Reader

OpenAI COO thinks AI for business is overhyped

theverge.com

24 pointsmfiguiere9 comments

"Human in the Loop" will be the major cooling factor of this hype cycle. Any feature proposal I hear always includes, "and of course there is a human in the loop to verify the output." The human needs the ability to create the content themselves if they hope to be effective verifiers. So overall you haven't really eliminated work, just transformed it. If its highly structured data like market reports or sporting results and can perform without verification, existing AIs have already been able to produce content like this for several years.

The only really high value prop I see for enterprise AI in the coming years is as a librarian. All companies have persistent knowledge base issues, usually stemming from multiple teams using multiple platforms. Having an assistant to comb through years of internal company documents and correlate information between them will be invaluable.

So, a search engine, then!

At my former company, they used a product to search across Google docs, confluence docs, help desk tickets, Amy source of tickets and documents they could get their hands on. It was actually quite useful!

Exactly. You still haven't eliminated the human in the loop if you make an AI librarian that generates "insights". Thanks to hallucination and the need to be able to back what you're saying up with something reproducible, you'll still need a list of citations you may have to check up on yourself. To wit, a search engine suffices. One exception might be if you are willing to tolerate inconsistency (or even lies) and aren't planning to share the result, just privately consume it with the appropriate level of caution.

Pretty much. Its just that most people are too lazy to look through results and synthesize information themselves. Also it would be helpful if the LLM could identify outdated or conflicting information, but I don't know if that is within their capabilities.

One human can manage far more work via AI than without, which still leads to job loss. Even if a human is still needed in the loop to verify the output.

e.g. how much faster can a language translator work with AI driven translation as a first pass, and human editing afterwards?

Clearly fewer people are required. And this is how innovations tend to go... manual book copying -> printing press and so on

It only leads to job loss if the labor demand is capped to the status quo.

Which will actually be the case in far less situations long term than people seem to think.

The reality is the early adopters who downsize staff to maintain current levels of output are shooting themselves in the foot as what's going to happen down the road is competitors stay staffed up but expand the scope of labor pulling away market share from the early adopters.

For example, you could cut staff and use AI to continue to produce mass market copy selling your products. Or you could keep the headcount the same and enhance each employee with AI in order to develop a pipeline spitting out copy optimized to the long tail of your customer base improving conversion in segments there was previously no way to tap with positive ROI.

i.e. Book publishers in the age of the printing press could have kept a roster of only a handful of authors and fired most of their transcribing staff or kept a large headcount for an expanded roster of authors. Scaling up productivity without scanning down labor should generally be the first place a wise company's mind goes, not scaling up productivity to scale down labor.

Individual jobs will shift (no need for transcribers with a printing press), but overall headcount shouldn't shift much as long as there's sufficient competition in a given market segment.

Yes, people get repurposed into new roles, but the jobs first in line for automation get downsized or eliminated.

Given that unemployment is close to all time lows, it's clear that technology doesn't increase unemployment in the aggregate or longer term. When real incomes rise, previously economically non viable jobs become viable.

But voice actors, translators, truckers will absolutely lose their jobs and unlikely to command the same compensation in whatever new venue they switch to.

At no point have I ever thought "man, what corporations really need is a better chatbot."

Chat as an I/O interface is very useful for AI companies, and the training data available by having many users interfacing with a LLM through natural language is again great for AI companies.

But it's one of the most boring and worthless applications of LLMs as a technology which is only impressive because of the unexpectedly significant advances of that underlying technology.

Yet the industry has bent over backwards to structure everything around the runaway success of the low hanging fruit.

In enterprise, the classifier role of LLMs is probably the immediate example of a smart and transformative application of the tech that could radically transform day to day operations. But ironically OpenAI's commitment to only supporting chat models and depreciating support for closer to pretrained text complete models means they'd be out of the running for the best fit for such a deployment in enterprise.

My hope for 2024 is that we finally see LLMs grow beyond the chatbot fad into their broader set of uses.

I just think that this is a manifestation of poor management. A responsible and competent executive would carefully research the actual capabilities and limitations of AI and not release a chat bot without testing it thoroughly with users.

So first you have to find a smart executive who is willing to dedicate time to research and set appropriate expectations and business goals given the actual capabilities of AI, and then you need to execute a software project very well involving this cutting edge technology.

Most executives do not perform step 1 or step 2 very well. Just because both are fairly involved tasks.

Reply on news.ycombinator.com