Forgot your password?
Close
typodupeerror
AI Businesses

Jeff Bezos' AI Startup Aims To Build an 'Artificial General Engineer' (theverge.com) 67

Posted by BeauHD from the more-details-emerge dept.
Jeff Bezos says his new AI startup, Prometheus, is working toward an "artificial general engineer" capable of helping design complex physical products such as robots, drugs, manufacturing systems, and rocket engines. The Verge reports: The NYT first reported on Prometheus last November, but now Bezos is sharing more information about the startup after a $12 billion funding round, putting the company at a $41 billion valuation. Bezos serves as co-CEO of Prometheus alongside Vik Bajaj, who co-founded Alphabet's health-focused research group, Verily. The startup currently has around 150 employees.

The tools Prometheus intends to build could help develop physical products across several industries, including robotics, drug design, and manufacturing, the NYT reports. "Blue Origin is a perfect example of a company that could benefit from the tools that Prometheus is building," Bezos tells the NYT. "Any company that is building sophisticated devices -- like rocket engines -- would benefit greatly from this kind of technology."
This discussion has been archived. No new comments can be posted.

Jeff Bezos' AI Startup Aims To Build an 'Artificial General Engineer' More

Jeff Bezos' AI Startup Aims To Build an 'Artificial General Engineer'

Comments Filter:

  • He should fix his old-space rocket company first (Score:2, Funny)

    by Anonymous Coward
    Musk is kicking his ass every way.
  • In my experience with AI tools, they generate a LOT of text that you have to read if you want to understand what it's trying to do. I can't imagine a group of managers reading through 3000 pages of AI output every day, which is what would have to happen to know if your 'AI Engineer' is doing stuff right.

    Otherwise, are you willing to spend billions on projects where you don't know what was actually done?
    • If you ask for a short summary, it can do that. What I would want it to do... is for me to explain in English what I want a board to do, what size I need it in, and other parameters. It can lay out a PCB, order the parts, Send the gerbers to PCB to a board shop, and have a working board in a week. Right now, an experienced Engineer should review the schematics and the layout, and give feedback to the AI, but there will come a day when that Engineer will be redundant the way things are going. As for s

      • Re: (Score:2)

        by hwstar ( 35834 )

        The lone wolf consultant probably won't be able to afford the service. Without it taking care of the gating items which require buy in from all stakeholders in a large corporation, it will output a sub-optimal design. The communication overhead to achieve buy-in and resolving problems with overcoming problems tool vendor software, and debugging, is what takes up most of an engineer's time today in a large company. AI which attacks this communication burden along with handling the boring and repetitive aspe

      • AI true believers always saying what it will do not what it can do.
        • Didn't you read the summary?

          Bezos tells the NYT. "Any company that is building sophisticated devices -- like rocket engines -- would benefit greatly from this kind of technology."

          This yet-uncreated-technology has perfectly understood capabilities, and will benefit you greatly.

      • Re: (Score:2)

        by dfghjk ( 711126 )

        If only AI could exaggerate like you do.

    • Otherwise, are you willing to spend billions on projects where you don't know what was actually done?

      Funny you should mention that [yahoo.com].

    • Re:AI generates a LOT of words that need to be rea (Score:5, Funny)

      by geekmux ( 1040042 ) on Friday June 12, 2026 @09:18PM (#66189520)

      I can't imagine a group of managers reading through 3000 pages of AI output every day..

      Any PHB worth their hair points knows the 'right' thing to do in that scenario is to pack all that output deep into the ChatGPT bong bowl and hit Enter, sparking the Flame of Delusion..

      ..which of course makes the executives cheer..

      • Otherwise, the executives will have to admit they are wrong, which is career suicide. If you admit you were wrong, another executive can use that admission to backstab you.

        • Otherwise, the executives will have to admit they are wrong, which is career suicide. If you admit you were wrong, another executive can use that admission to backstab you.

          When "another" executive turns out to be AI, and all the meatsack execs turn on themselves?

          The fucking popcorn, will be pay-per-kernel.

    • Re: (Score:2)

      by MrKaos ( 858439 )

      I can't imagine a group of managers reading through 3000 pages of AI output every day,

      I dunno, it could be kind of funny watching them trying to get a machine to make sense of reality.

    • Reading 3000 pages of AI output? To overextend the metonym, Is that better or worse than having your liver eaten by an eagle every day?

    • Re: (Score:2)

      by dfghjk ( 711126 )

      There's an AI that will help with that reading, and yes, people will be willing to spend billions when it is not their billions. Plus, other billionaires are telling them what AI will "actually do", even though they are lying.

    • The current AI tools are Large Language Models (LLMs), pretty much just advanced chatbots. These are neural networks which process and generate texts. This ability is not really helpful for an "Artificial Engineer". This has to be a different kind of neural network, one that processes formulas, design ideas and numbers. Also a simulation module certainly helps.

      The current AIs can do more than just processing words, because the text generator is used as an interface to other software- and ai modules. E.g.
      • Go ask ChatGPT or Claude a question no one has ever written down. How many hamsters would it take to chew through Africa like a locust swarm? If London Bridge were built out of paper, where would it first structurally fail? The LLM will probably be able to either answer correctly or take a damn good shot at it. This is evidence that LLMs can at least approach engineering problems. Same for asking them for a circuit layout for basic arbitrary tasks, as long as the answer is within context window.

    • Re: (Score:2)

      by allo ( 1728082 )

      AI follows instructions. You can
      a) Ask it to be brief.
      b) Let it babble (this can help it to work better as it doesn't have parallel "thoughts" but can only think by generating text) and ask it for a summary afterward

    • Parent should have been the FP and I'm glad to see it spanned about half the discussion. But I'm just judging by the scrollbar. Mostly I only check the longer discussions on Slashdot for Funny these days. Saves vast amounts of time while producing the same amounts of disappointment.

      However on the verbosity topic, I actually ran the obvious test past a number of genAIs. "Can you answer any question as briefly as possible?" Rather to my surprise, about half of them managed to answer "Yes" and then shut up.

      But

  • And who will be the accountability sink (Score:4, Interesting)

    by hwstar ( 35834 ) on Friday June 12, 2026 @07:23PM (#66189422)

    when the bridge or building collapses, or many die due to a flaw in a drug developed by these AGE's?

    I bet it won't be anyone in this new startup due to the corporate veil.

    • The company will be, not an individual

    • Re:And who will be the accountability sink (Score:4, Insightful)

      by Inglix the Mad ( 576601 ) on Saturday June 13, 2026 @01:55AM (#66189710)
      This is similar to the radiology argument.

      Company builds "AI" (it isn't AI but let's pretend) that can "read" images. Company goes to sell this to a hospital CEO. They say it will do the work that radiologists would do, and it only costs (X) tokens per scan. Imagine the savings! Hospital asks about liability, and here's the kicker: The AI company says just have one of the remaining human doctors the hospital has on-site to review the scan.

      Now skipping the part where it says it can do the work, it's dumping liability on the human who now not only has to review their own work but also the AI. The AI doesn't have to be perfect... the human has to be perfect. In a logical world the CEO would tell the company to go sell their hokum elsewhere until they're willing to put their own reputation AND finances on the line to stand behind their work. However if the average CEO can sign a contract to pay the company 15% less than a human overall, then add that 15% to the bottom line, they'll get a bonus, so you'd best get to work doc.

      When the mistake happens, you'll get sued... instead of the "AI" system and it just sucks to be you person who went to school to become an actual SME. We've already seen them blame software devs/engineers when AI writes crappy code that causes problems. You think they aren't going to dump other bad crap on humans when the tool breaks?
  • You want to make the genie that makes whatever you want.

  • Based on AI flaws so far... (Score:3)

    by PhantomHarlock ( 189617 ) on Friday June 12, 2026 @08:00PM (#66189462)

    You'll end up with the worst employee you've ever had. A narcissist who sounds completely compelling but is completely wrong, or just wrong enough that it sounds right, but the load calculation is off by a small factor, no one else catches it and the bridge fails under a certain condition, someone dies.

    There's no intelligence when it's just mindlessly trying to slot the right word in the next position. I realize specialized AIs are starting to have some particular skills, but it still seems so untrustworthy that you still need intensive design reviews by senior engineers, assuming the AI engineer is an idiot and needs double checking at every turn.

    • Re:Based on AI flaws so far... (Score:4, Informative)

      by geekmux ( 1040042 ) on Saturday June 13, 2026 @09:50AM (#66189982)

      The deadly part, isn't when well-trained intelligent humans can call AI on it's bullshit.

      The deadly part, is when all those humans die off and get replaced with the drug-addled screen addicted generation that doesn't even care to anymore.

      When apathy dies, humanity dies.

    • Re: (Score:2)

      by booch ( 4157 )

      sounds completely compelling

      That's not how engineers use AI agents any more. We don't just look at what it says — we have it prove that it's doing the right thing. We just look at the proof, so we don't have to look at all the details, nor blindly believe what they AI says. We also have adversarial reviews, which are typically more thorough (but less experience-based) than human reviews.

  • Drug discovery part is BS (Score:5, Informative)

    by methano ( 519830 ) on Friday June 12, 2026 @08:11PM (#66189470)
    I'm an organic chemist. I've worked in drug discovery for 45 years. It's hard. Jeff Bezos ain't nowhere close to figuring it out. I don't know about the other parts, but I know for sure the drug discovery stuff is bullshit.
    • Well then, you must know better than most people that when you are on the right drugs it not only seems possible, but easy!

    • Re: (Score:2)

      by dfghjk ( 711126 )

      But Jeff Bezos doesn't care about figuring it out, he only cares about owning a replacement for you. Once you are gone, there is no one that can expose that he didn't figure it out.

  • Why does he need to do that? Just in the other thread I've been sternly assured these single men *must* have their hundreds of billions of dollars so they can fund these new ventures, otherwise how on Earth would they do it?!?! Go out and seek other people with money to put towards it?! Well at rate we may as well be commies amirite?!

    • Why would he spend his own money?

      • Exactly, so why does anyone need hundreds of billions of dollars of it? Tax that shit into the ground.

        1000 millionaires will make wiser investment choices than one billionaire. After all, that's a bedrock principle of capitalism, the market is wise.

        • Re: (Score:2)

          by dfghjk ( 711126 )

          There is nothing wise about millionaires or billionaires, they are the toxic part of capitalism. Monopolizing capital destroys free markets.

      • Exactly. You get rich by convincing others to spend their money.

  • Since ChatGPT3 we've left behind the financial reality of stocks being worth what their expected returns are over time, so either the company pays dividends or gets bought and people that own the stock get actual money back for the money they paid in, and we've now reaching the stage of orbit where everything is just a complete fantasy designed to goose more money out of people that somehow haven't run out spare cash to throw at stocks yet.

    AI hasn't replaced coders, despite it being declared so. AI has tem
    • Might not have replaced coders but it has completely changed the job. Used to be you had to figure out what you wanted to do, then how, then do it, then tests. Now you just need the "what". The AI figures out the rest. Iterate a bit if you want changes but it's still much faster. Have an idea, prototype it in 10 minutes instead of a week.
  • Well, if I were prompting the LLM, I would simply request that it not make any mistakes.

  • I can see this. (Score:3)

    by gurps_npc ( 621217 ) on Friday June 12, 2026 @09:59PM (#66189550) Homepage

    Most of the proposed AI projects sound foolish to me.

    But I think this could work. The trick is NOT to have gigantic expectations.

    Forget about 'Design me a bridge'.

    Instead think "Create a table of how much weight can a bridge Suspension bridge spanning 1 mile hold, by different construction materials."

     

  • seems dumb (Score:3)

    by OrangeTide ( 124937 ) on Friday June 12, 2026 @10:08PM (#66189558) Homepage Journal

    Go for the low hanging fruit, a position that is easy to automate and consumes the largest paycheck: MBA

    • It's been happening [cnbc.com]:

      "The trend is expected to continue into 2026: One in five (20%) businesses are expected to use AI to flatten their organizational structure, slashing over half of current middle management positions"

      I don't think AI deserves the blame, though.

      • Replacing worthless people with worthless software that boils the oceans. It's not exactly progress, more of a lateral move on the stupidity track of humanity.

    • Re: (Score:2)

      by dfghjk ( 711126 )

      Republicans.

  • Seriously, how disconnected are these people?

  • I'm rather doubtful of the plan's viability; but if they are willing to justify the company name by chaining Bezos to a rock so a bird can eat his liver repeatedly I could be persuaded.

  • Given how engineering is a registered profession in many places in the world, I'd like to see gain its professional accreditation.

    • Re: (Score:2)

      by dfghjk ( 711126 )

      That's just playing games with words. In places where there are professional Engineers, you can be an engineer without accreditation, you just can't be an Engineer. The government and private organizations don't own words, they control titles.

      Besides, accreditation doesn't need to be "gained" when it's an AI, just hang the label on it and threaten lawsuits for those who object. It's all about power and money. Earning it is for you, not them.

      • But the government also sets other rules. You can't operate on people in a hospital of you're not a real, accredited surgeon. You can't get the contract to build a railroad bridge unless you're a real, accredited engineer.

        It's not just words, the whole point is to ensure that those that need a service have a way of knowing that the person offering it has at least been vetted somehow. Good luck getting a job without that accreditation in an industry where it exists

Slashdot Top Deals

* * * * * THIS TERMINAL IS IN USE * * * * *

Close