Oh, come on! You just KNEW it was coming.

OpenAI is exploring "salaries" for agents.


__________________________


Project Counsel Media is a division of Luminative Media. We cover the areas of cyber security, digital technology, legal technology, media, and mobile technology.


About Luminative Media: our intention is to delve deeper into issues, at greater length and with more historical and social context, in order to illuminate pathways of thought that are not possible to pursue through the immediacy of daily media. For more on our vision please click on our logo:


________________


Yes, hallucinations are still embedded in LLMs, but do carry on.


______________________________________________


BY:


Eric De Grasse

Chief Technology Officer


Member of the Luminative Media / Project Counsel Media teams



______________________________________________________


10 March 2025 (New York, NY) -- OpenAI is apparently considering tiered pricing for Deep Research style agents: $2,000/month for work equivalent to "high income knowledge workers", and up to $20,000/month for "PhD-level research agents".


Setting aside general amusement at the comparison with actual PhD incomes, Deep Research today consistently makes outrageous factual mistakes that would be unforgivable in a summer intern (he/she would be fired). As I have written before, hallucinations are embedded in LLMs, and not going away.


More fundamentally, presuming that does gets solved (and today nobody thinks it will), it seems very unlikely that automation will cost more than people rather than less, especially if (see related story below) there aren’t really any moats.


This week’s buzzy AI thing (do keep up people!) is Manus, which launched a very cool demo of an AI agent that can apparently process a bunch of complex multi-stage tasks, including "Operator"-like use of third-party websites.


It’s based (or at least domiciled in) Singapore, without much clarity about backers, but there are some suggestions that under the hood it’s really Anthropic’s Claude with a bunch of other tools bolted together.


Meanwhile, the demo use-cases are indeed cool, but they remind me painfully of Rabbit in their breezy and slightly suspicious over-simplification - "arrange a two month holiday in five countries!"


If the demo is real, though, and if it’s their model, then this is a little bit of another DeepSeek - in the sense that it demonstrates the increasing commodification of this technology while the real use-cases are built elsewhere. Because in tech, everything eventually gets commoditized.


And speaking of hallucinations, after, what, $1 trillion of investment in LLMs, we get ....


* * * * * * * * * * * * * * * 


For the URL link to this post, please click here


For our other recent posts, please click here



* * * * * * * * * * * * * * *