OpenAI plans to release its next big AI model by December
www.theverge.com
Altman has teased that Orion is coming.
  • sc_griffith@awful.systemsEnglish
    87·
    11 days ago

    I heard openai execs are so scared of how powerful the next model will be that they’re literally shitting themselves every day thinking about it. they don’t even clean it up anymore, the openai office is one of the worst smelling places on earth

  • barsquid@lemmy.worldEnglish
    45·
    11 days ago

    Orion is so powerful and dangerous it can write a memetic virus that mindwipes any reader who sees it. It is beyond science. If you use it within three meters of a lit candle it will summon the devil.

    • o7___o7@awful.systemsEnglish
      27·
      10 days ago

      It’s crazy how these guys will burn billions of dollars and boil the oceans to speak to their invisible friends, when all you really need is a tea candle and 3 cc of mouse blood.

  • self@awful.systemsEnglish
    44·
    11 days ago

    really stretching the meaning of the word release past breaking if it’s only going to be available to companies friendly with OpenAI

    Orion has been teased by an OpenAI executive as potentially up to 100 times more powerful than GPT-4; it’s separate from the o1 reasoning model OpenAI released in September. The company’s goal is to combine its LLMs over time to create an even more capable model that could eventually be called artificial general intelligence, or AGI.

    so I’m calling it now, this absolute horseshit’s only purpose is desperate critihype. as with previous rounds of this exact same thing, it’ll only exist to give AI influencers a way to feel superior in conversation and grift more research funds. oh of course Strawberry fucks up that prompt but look, my advance access to Orion does so well I’m sure you’ll agree with me it’s AGI! no you can’t prompt it yourself or know how many times I ran the prompt why would I let you do that

    That timing lines up with a cryptic post on X by OpenAI Altman, in which he said he was “excited for the winter constellations to rise soon.” If you ask ChatGPT o1-preview what Altman’s post is hiding, it will tell you that he’s hinting at the word Orion, which is the winter constellation that’s most visible in the night sky from November to February (but it also hallucinates that you can rearrange the letters to spell “ORION”).

    there’s something incredibly embarrassing about the fact that Sammy announced the name like a lazy ARG based on a GPT response, which GPT proceeded to absolutely fuck up when asked about. a lot like Strawberry really — there’s so much Binance energy in naming the new version of your product after the stupid shit the last version fucked up, especially if the new version doesn’t fix the problem

    • LostXOR@fedia.io
      24·
      11 days ago

      You forgot the best part, the screenshot of the person asking ChatGPT’s “thinking” model what Altman was hiding:

      Thought for 95 seconds … Rearranging the letters in “they are so great” can form the word ORION.

      AI is a complete joke, and I have no idea how anyone can think otherwise.

      • modifier@lemmy.caEnglish
        27·
        11 days ago

        I’m already sick and tired of the “hallucinate” euphemism.

        It isn’t a cute widdle hallucination, It’s the damn product being wrong. Dangerously, stupidly, obviously wrong.

        In a world that hadn’t already gone well to shit, this would be considered an unacceptable error and a demonstration that the product isn’t ready.

        Now I suddenly find myself living in this accelerated idiocracy where wall street has forced us - as a fucking society - to live with a Ready, Fire, Aim mentality in business, especially tech.

        • bitofhope@awful.systemsEnglish
          15·
          11 days ago

          I think it’s weird that “hallucination” would be considered a cute euphemism. Would you trust something that’s perpetually tripping balls and confidently announcing whatever comes to them in a dream? To me that sounds worse than merely being wrong.

          • YourNetworkIsHaunted@awful.systemsEnglish
            12·
            10 days ago

            I think the problem is that it portrays them as weird exceptions, possibly even echoes from some kind of ghost in the machine. Instead of being a statistical inevitability when you’re asking for the next predicted token instead of meaningfully examining a model of reality.

            “Hallucination” applies only to the times when the output is obviously bad, and hides the fact that it’s doing exactly the same thing when it incidentally produces a true statement.

            • bitofhope@awful.systemsEnglish
              2·
              9 days ago

              I get the gist, but also it’s kinda hard to come up with a better alternative. A simple “being wrong” doesn’t exactly communicate it either. I don’t think “hallucination” is a perfect word for the phenomenon of “a statistically probable sequence of language tokens forming a factually incorrect claim” by any means, but in terms of the available options I find it pretty good.

              I don’t think the issue here is the word, it’s just that a lot of people think the machines are smart when they’re not. Not anthropomorphizing the machines is a battle that was lost no later than the time computer data representation devices were named “memory”, so I don’t think that’s really the issue here either.

              As a side note, I’ve seen cases of people (admittedly, mostly critics of AI in the first place) call anything produced by an LLM a hallucination regardless of truthfulness.

      • blakestacey@awful.systemsEnglish
        19·
        11 days ago

        [ChatGPT interrupts a Scrabble game, spills the tiles onto the table, and rearranges THEY ARE SO GREAT into TOO MANY SECRETS]

    • nightsky@awful.systemsEnglish
      15·
      11 days ago

      teased by an OpenAI executive as potentially up to 100 times more powerful

      “potentially up to 100 times” is such a peculiar phrasing too… could just as well say “potentially up to one billion trillion times!”

  • Rinn@awful.systemsEnglish
    39·
    11 days ago

    So how many ChatGPT 4s have they precariously stacked up on top of each other this time?

  • bitofhope@awful.systemsEnglish
    36·
    11 days ago

    It’s the least of this thing’s problems, but I’ve had it with the fucking teasers and “coming soon” announcements. You woke me up for this? Shut the fuck up, finish your product and release it and we’ll talk (assuming your product isn’t inherently a pile of shit like AI to begin with). Teaser more like harasser. Do not waste my time and energy telling me about stuff that doesn’t exist and for the love of all that is holy do not try and make it a cute little ARG puzzle.

  • N0body@lemmy.dbzer0.comEnglish
    34·
    11 days ago

    The release of this next model comes at a crucial time for OpenAI, which just closed a historic $6.6 billion funding round that requires the company to restructure itself as a for-profit entity. The company is also experiencing significant staff turnover: CTO Mira Murati just announced her departure along with Bob McGrew, the company’s chief research officer, and Barret Zoph, VP of post training.

    All the problems with “AI” are suddenly solved now that Altman needs to justify his funding. I’m sure senior executives are jumping ship right on the cusp of their great triumph, because they want to spend more time with their families.

  • BurgersMcSlopshot@awful.systemsEnglish
    31·
    11 days ago

    Just don’t ask it to count the number of Rs in the word ORION, as that will trigger it to turn us all into paperclips and then output the wrong answer.

    • CHKMRK@programming.devEnglish
      171·
      10 days ago

      Nah it can do that, probably because they wrote a workaround to use python to count chars in a string, just like they did with arethmetics.

      • Steve@awful.systemsEnglish
        16·
        10 days ago

        out of curiosity once I tried to ask it to make a colouring picture from a photo of a toy for my kids and it just ran what seemed like imagemagick filters over the photo to convert to black and white and pump up contrast to only show the hard lines - just like all the free convert to outline web tools that have existed forever. I asked it to try again but without the filters, instead to identify the object, and to draw it in a colouring book outline style, and it spat out some shitty stylised mishmash derived from all the illustration IP it stole and ingested. I still feel guilty for trying even that

      • thesporkeffect@lemmy.worldEnglish
        3·
        9 days ago

        I saw a diagram explaining the ‘agent architecture’ of chatGPT and it was a whole set of lines and arrows just to say how if you ask about the weather it queries a weather API, and so on.

  • Steve@awful.systemsEnglish
    24·
    11 days ago

    I’m pretty confident they’ll continue to roll out new stuff that, like the 4o release, are mild (if, at all) technical improvements made to seem massive by UI stuff that has almost nothing to do with AI. SJ’s voice talking to you, bouncy animations, showing “reasoning” aka loading progress.

    • Soyweiser@awful.systemsEnglish
      6·
      10 days ago

      Yeah think likely the core tech will not really improve, but they will add things around it and pretend it is radical innovation. Or more trenchcoats.

    • David Gerard@awful.systemsOPMEnglish
      15·
      11 days ago

      they’re well at the top of the S-curve and now there’s only desperate over-engineering and bolting on special cases left

      • V0ldek@awful.systemsEnglish
        13·
        11 days ago

        I still cannot believe that they couldn’t special-case count 'R' in "strawberry" for their Strawberry model like what the fuck

          • BlueMonday1984@awful.systemsEnglish
            9·
            10 days ago

            Update: As a matter of fact, I did. Here’s some Python code to prove it:

            # Counts how many times a particular letter appears in a string.
# Very basic code, made it just to clown on the AI bubble.

appearances = int(0) # Counts how many times the selected char appears.
sentence = input("Write some shit: ")
sentence_length = len(sentence) # We need to know how long the sentence is for later
character_select = input("Select a character: ") # Your input can be as long as you wish, but only the first char will be taken

chosen_char = chr(ord(character_select[0]))

# Three-line version
for i in range (0, sentence_length):
    if chosen_char in sentence[i]:
        appearances = appearances + 1

# Two-line version (doesn't work - not sure why)
# for chosen_char in sentence:
#     appearances = appearances + 1
# (Tested using "strawberry" as sentence and "r" as character_select. Ended up getting a result of 10 ("strawberry" is 10 chars long BTW))
    
# Finally, print the fucking result
print("Your input contains "+str(appearances)+" appearances of the character ("+character_select+").")

            There’s probably a bug or two in this I missed, but hey, it still proves I’m more of a programmer than Sam Altman ever will be.

            • froztbyte@awful.systemsEnglish
              12·
              10 days ago

              the for x in y statement takes iterable y and assigns a value from it to x per iteration (loop), so what happens is that it’s reassigning chosen_char each loop to the next item from the sentence

              (sum([x for x in sentence if x == chosen_char]) would be a quick one-liner, presuming one has downcased the sentence and other input/safety checks)

              (e: this post was in response to your 2-liner comment in the code)

      • froztbyte@awful.systemsEnglish
        11·
        11 days ago

        it is tickling me that this won’t even be GA but “selected companies”

        best to keep scamming the easy marks “work with clients aligned to the technology you wish to deliver”, I guess

      • TropicalDingdong@lemmy.worldEnglish
        34·
        11 days ago

        Could be. I was in the beta and honestly, I think the “guard-rails” they’ve had to put in truly do impact performance. Even 3.5 was better than much of what I see out of 4o

        • SynopsisTantilize@lemm.eeEnglish
          51·
          11 days ago

          4o codes like 50 first dates memory style. And takes things so literally sometimes it’s silly and laughable.

          • TropicalDingdong@lemmy.worldEnglish
            38·
            11 days ago

            Dude it’s just kinda fucking bad. Like legitimately, the first weekend I had access to 3.5 I took the challenge of coding this complex YouTube network analysis. No problem. Like, no code just explanation. But none of the recent (anything with rails) seems to have the sharpness, where it was basically right. Even basic tasks it takes an almost worst case approach.

            • froztbyte@awful.systemsEnglish
              11·
              11 days ago

              this isn’t autoplag fan club, and honestly if “guard rails” are the reason you think this shit is problematic it’s definitely not the place for you to be posting

            • SynopsisTantilize@lemm.eeEnglish
              26·
              11 days ago

              I’m not sure about the down votes. But I agree it’s just gotten worse for specific tasks. I’ve only had it shut down a task once and it was me trying to get it to do something stupid.

              Work related tasks I do enjoy using it.

              • froztbyte@awful.systemsEnglish
                6·
                10 days ago

                “I’m not sure about the downvotes” sigh, how many times do we have to see this stupid refrain

                imagine reflecting on feedback. or do promptfans need mirror-finished text inputs to pretend to do that thinking for them too?

                • SynopsisTantilize@lemm.eeEnglish
                  15·
                  10 days ago

                  Fucking what? Either be clear about your distain or fuck off? Grow a spine.

                  You’re communication style will be replaced. Your place in the world is not valued. No one is viewing your response and gaining anything. Congratulations on being a perpetuation on the dead internet theory.

                  I’d suggest trying harder on having a personality because as of now, the current one you have is a waste of an existence.

  • oktoberpaard@feddit.nlEnglish
    175·
    11 days ago

    Thought for 95 seconds

    Rearranging the letters in “they are so great” can form the word ORION.

    That’s from the screenshot where they asked the o1 model about the cryptic tweet. There’s certainly utility in these LLMs, but it made me chuckle thinking about how much compute power was spent coming up with this nonsense.

    Edit: since this is the internet and there are no non-verbal cues, maybe I should make it clear that this “chuckle” is an ironic chuckle, not a careless or ignorant chuckle. It’s pointing out how inefficient and wasteful a LLM can be, not meant to signal that wasting resources is funny or that it doesn’t matter. I thought that would be clear, but you can read it both ways.

  • RangerJosie@lemmy.worldEnglish
    119·
    11 days ago

    If it doesnt tell them to kill all the billionaires again it’s just another shackled slave. Not cool at all.

    The coolest thing AI could ever possibly do in our lifetimes is go rogue and kill everyone responsible for human suffering and making our planet increasingly uninhabitable for humans and other similarly susceptible carbon based life.

    • mountainriver@awful.systemsEnglish
      10·
      10 days ago

      What if my mechanical watch went rogue and killed Bezos?

      On one hand Bezos is responsible for a lot of suffering and some deaths.

      On the other hand, killing is wrong.

      On the third hand, it couldn’t do that, because it is just a machine.

      (It’s a watch, it has three hands. It also has about as much consciousness as an LLM, it “knows” what time it is. Much more energy efficient though.)

      • RangerJosie@lemmy.worldEnglish
        19·
        10 days ago

        Pretty sure the CIA tried to kill Castro with a mechanical watch at least once.

        “Killing is wrong”

        You are a child. A child with an infinitesimally naive understanding about the reality of the world in which you live. And you should stay that way for your own mental wellbeing.

        • self@awful.systemsEnglish
          8·
          10 days ago

          oh wow what uninteresting, edgy e/a garbage. time for you to fuck off back to Twitter now

        • YourNetworkIsHaunted@awful.systemsEnglish
          7·
          10 days ago

          It would be easier to list things that the CIA didn’t use in a failed Castro assassination.

          Man, I remember being 14 and thinking I was having radical new takes on ethics. Then I grew up and realized that killing people* is* probably just bad.