OpenAI: Out of Ideas (opens in new tab)

I saw an interesting quote from the deekseek CEO:

"When tackling a task, experienced people might reflexively tell you 'this is how it should be done,' while those without experience will experiment repeatedly, think deeply about the best approach, and ultimately find a solution tailored to the actual circumstances."

I do not think that OpenAI is out of ideas, but I do believe that large businesses have difficulty overcoming the innovator's dilemma. In addition to big businesses, there are also many small and very smart companies that are using investors' over exuberance to their advantage. Nevertheless, there are enough people trying that some are likely to find a way to develop higher-quality models at a lower cost.

> They’d rather give Sam Altman $200 per month for access to the world’s most mediocre researcher than get good results from a human.

Really going to need a citation for the claim that Deep Research is the world's "most mediocre researcher", the product was launched like 12 hours ago...

I'd buy it's not the best researcher in the world, but I'm willing to bet a lot that it's very far from the worst human researcher. I'd wager it's well above average.

I've felt this way since many of the other modalities they've announced to great fanfare fell by the wayside. Sora proved underwhelming. Likewise for voice mode. DALLE is far behind open alternatives.

It's hard to know because they're closed source, but evidence of advances in architecture is lacking.

OpenAI is not relevant in a world where DeepSeek is open source.

The post complains about token cost and invisible tokens, but the feature is only available to subscribers who pay a flat rate (not per token).

This article is one of the dumbest takes I have ever seen. The author claims that OpenAI is out of ideas because they are only improving what they already have since GPT3.

That's like saying, getting a new computer is pointless, because it only has a better CPU, more memory and a better GPU, but is still the same parts, so it's pointless and they are out of ideas for making new computers.

My hot take is that these "reasoning" models (o3/R1, et. al.) are just silly stop-gap tricks. Any kind of agentic system built on foundational models is more than capable of outputting equal (or higher quality with tool usage) results. They are just shipping a highly generic version of that under the hood, and you're better off just custom tailoring an agentic process for your specific workflow rather than relying on theirs. There's a good reason Anthropic hasn't bothered coming out with one. I suspect these types of models will become a dead end as the foundational ones advance and agentic patterns become commonplace.

The author of the article is not an expert in this field. You should give him no more credence than you would give to, say, a plumber talking about astrophysics. Or astrpophysicist talking about plumbing.

'The truth is they plateaued long ago with GPT-3'

Hmmm....