4 ms·

What I’m afraid of is that he and Ilya are not as good and smart as they paint themselves. And that a lot of key people had left (i.e. to Anthropic). And that

What I’m afraid of is that he and Ilya are not as good and smart as they paint themselves. And that a lot of key people had left (i.e. to Anthropic). And that by pure inertia they have GPT-5 on their hands and not much control over where this technology is going. I can’t tell for certain, but it does look like one of their corner pieces, the ChatGPT system prompt which sits at the funnel of the data collection had degraded significantly from the previous version. Had the person that was the key to the previous design left? Or it no longer matters? One could argue that OpenAI is very hot and everyone would want to work there. But a lot of newcomers only create more pressure for the key people. And then there is the inevitable leakage problem.

bmitc 4y ago
> What I’m afraid of is that he and Ilya are not as good and smart as they paint themselves. This describes almost any venture capitalist or high-profile startup founders, as far as I can tell. Most don't realize their either privileged path or lucky path or both had more to do with it than their smarts. I really like James Simons as he mostly attributes his success to luck and being able to hire and organize smart people and give them the tools they need to work. He basically describes it as luck and taste, despite his actual smarts and his enormous impact on the world.
- saghm 4y ago
  > I really like James Simons as he mostly attributes his success to luck and being able to hire and organize smart people and give them the tools they need to work. He basically describes it as luck and taste, despite his actual smarts and his enormous impact on the world. Plenty of really smart people don't end up having a big impact on the world, and it's possible to make a difference without being an outlier in terms of intelligence. Everyone who has made an impact has benefited to some degree by circumstances beyond their control though, so even if someone is genuinely smarter than anyone else, it's a fallacy for them to assume that it was the determining factor in their success and a guarantee of future success.
- iibarea 4y ago
  ... and Simons would maybe be the most justified in overlooking luck, but he's smart enough to realize how random the world is. Peter Norvig also emphasizes the role of luck in his life. It's honestly a very good test of self-awareness and empathy, though there's def some negative selection against those traits in sv.
- Mistletoe 4y ago
  I don’t know everything about him but from what I do know, I would put Bezos in the “not just luck” very lonely camp. I think his Day 1 work and iterate every day idea is just that powerful and real and he really did it instead of talking about it. Even though he says he won several lotteries to get where he is, I’m not so sure.
sebzim4500 4y ago
I'm sure they are both pretty smart, but if anything that makes their apparent monopoly more concerning.
- dnissley 4y ago
  Monopoly over what?
  - sebzim4500 4y ago
    LLMs that actually work. They are on GPT-4 and no one else is close to GPT-3.5.
    - shagie 4y ago
      If no one else has yet to create the same quality of product, is that really a monopoly? Does a given chef have a monopoly on their signature dish? It feels like people are tossing around "monopoly" whenever it feels like there's a company that has produced a quality product and people want to hobble it because no one else has committed the resources to producing something comparable.
      - sebzim4500 4y ago
        I don't want to hobble it at all. To the contrary, I hope the competition pulls their heads out of their asses.
FormerBandmate 4y ago
You can use Anthropic’s chatbot in Quora’s Poe app. Right now it isn’t as good as Bing or ChatGPT. Misses some basic logic things, and the “As an AI language model” BS still stops it from doing fun things like making Jesus rap battle Gus Fring (that was like a month ago, someone in the replies got it to do that so I’ll have to check it out again). I’d have to see how it is at writing PowerPoints but idk
- IncRnd 4y ago
  Verse 1 - Jesus I'm the son of God, the King of Kings You're just a drug lord, selling crystal meth and things My teachings change lives, bring peace to the world You bring addiction, violence and pain, unfurled Chorus Jesus, the savior, the light in the dark Gus Fring, the villain, who leaves his mark Verse 2 - Gus Fring You talk a big game, but where's your proof? I've built an empire, with power that's bulletproof Your miracles are outdated, my tactics are new I'll take you down, no matter what you do Chorus Jesus, the savior, the light in the dark Gus Fring, the villain, who leaves his mark Verse 3 - Jesus My love conquers all, it's the greatest force Your money and power, just lead to remorse You're just a man, with a fragile ego I'll show you mercy, but you reap what you sow Chorus Jesus, the savior, the light in the dark Gus Fring, the villain, who leaves his mark Verse 4 - Gus Fring You may have won this battle, but the war is not done I'll continue to rise, until I've won You may have followers, but they'll never be mine I'll always come out on top, every time Chorus Jesus, the savior, the light in the dark Gus Fring, the villain, who leaves his mark Outro In the end, it's clear to see Jesus brings hope and love, for you and me Gus Fring may have power, but it's not enough Jesus is the way, the truth, the life, and that's tough.
  - FormerBandmate 4y ago
    Huh, last time I checked it it gave me a message about how that was “offensive to Christians”. I’ll have to check it out again
    - shagie 4y ago
      While I'm not the person who did this bit - there's a difference between using the API directly and using ChatGPT. The API doesn't get run through the moderation model to check if you're asking for acceptable things or if the model is getting into 'unsafe' areas. Fire up the playground ( https://platform.openai.com/playground https://platform.openai.com/playground ) and construct the prompt.
sroussey 4y ago
> ChatGPT system prompt which sits at the funnel of the data collection had degraded significantly from the previous version They purposely moved free users to a simpler/cheaper model. Depending on your setting and if you are paying, there are three models you might be inferencing with.
- startupsfail 4y ago
  I’m not talking about GPT-3.5 vs GPT-4. I’m talking about a change to their system prompt.
jstummbillig 4y ago
There are some vague ideas and fears here. Understandable. Trying find a silver lining from where to get somewhere: Where would GPT4 and onwards be better housed? Is there a setup -- an individual, a company, an institution, a concept, license -- where the whole thing would clearly better fit, than with OpenAI? Note, I am not suggesting that they are particularly un/qualified or un/trustworthy. I am just trying to figure, if the problem is with the nature of the technology, that maybe there is not entity or setup, that would obviously be a good fit for governing gpt because gpt is simply scary, or this is a personality issue.
- startupsfail 4y ago
  I think that a mix of public-private ownership is likely a good idea. And there should be some serious oversight. The decisions at the level of proliferation of plutonium and building atomic bombs should not be done solely by a startup founder under a pressure to deliver, to keep the hype, to not let the team be headhunted, etc. I also don’t know about your familiarity with Ycombinator, but they are successful partially because they are pretty brutal. They are not the peace-time CEOs. And I’m not sure, if you want the AGI to be developed in a ways of a war-time CEO. And this is exactly what is going on now. I’d probably call for organizing a consortium of US-Government-Microsoft-Google-Intel-Nvidia-OpenAI to lead the decision making process and to relieve the pressure on OpenAI to some degree.

bmitc 4y ago

> What I’m afraid of is that he and Ilya are not as good and smart as they paint themselves. This describes almost any venture capitalist or high-profile startup founders, as far as I can tell. Most don't realize their either privileged path or lucky path or both had more to do with it than their smarts. I really like James Simons as he mostly attributes his success to luck and being able to hire and organize smart people and give them the tools they need to work. He basically describes it as luck and taste, despite his actual smarts and his enormous impact on the world.

saghm 4y ago
> I really like James Simons as he mostly attributes his success to luck and being able to hire and organize smart people and give them the tools they need to work. He basically describes it as luck and taste, despite his actual smarts and his enormous impact on the world. Plenty of really smart people don't end up having a big impact on the world, and it's possible to make a difference without being an outlier in terms of intelligence. Everyone who has made an impact has benefited to some degree by circumstances beyond their control though, so even if someone is genuinely smarter than anyone else, it's a fallacy for them to assume that it was the determining factor in their success and a guarantee of future success.
iibarea 4y ago
... and Simons would maybe be the most justified in overlooking luck, but he's smart enough to realize how random the world is. Peter Norvig also emphasizes the role of luck in his life. It's honestly a very good test of self-awareness and empathy, though there's def some negative selection against those traits in sv.
Mistletoe 4y ago
I don’t know everything about him but from what I do know, I would put Bezos in the “not just luck” very lonely camp. I think his Day 1 work and iterate every day idea is just that powerful and real and he really did it instead of talking about it. Even though he says he won several lotteries to get where he is, I’m not so sure.

sebzim4500 4y ago

I'm sure they are both pretty smart, but if anything that makes their apparent monopoly more concerning.

dnissley 4y ago
Monopoly over what?
- sebzim4500 4y ago
  LLMs that actually work. They are on GPT-4 and no one else is close to GPT-3.5.
  - shagie 4y ago
    If no one else has yet to create the same quality of product, is that really a monopoly? Does a given chef have a monopoly on their signature dish? It feels like people are tossing around "monopoly" whenever it feels like there's a company that has produced a quality product and people want to hobble it because no one else has committed the resources to producing something comparable.
    - sebzim4500 4y ago
      I don't want to hobble it at all. To the contrary, I hope the competition pulls their heads out of their asses.

FormerBandmate 4y ago

You can use Anthropic’s chatbot in Quora’s Poe app. Right now it isn’t as good as Bing or ChatGPT. Misses some basic logic things, and the “As an AI language model” BS still stops it from doing fun things like making Jesus rap battle Gus Fring (that was like a month ago, someone in the replies got it to do that so I’ll have to check it out again). I’d have to see how it is at writing PowerPoints but idk

IncRnd 4y ago
Verse 1 - Jesus I'm the son of God, the King of Kings You're just a drug lord, selling crystal meth and things My teachings change lives, bring peace to the world You bring addiction, violence and pain, unfurled Chorus Jesus, the savior, the light in the dark Gus Fring, the villain, who leaves his mark Verse 2 - Gus Fring You talk a big game, but where's your proof? I've built an empire, with power that's bulletproof Your miracles are outdated, my tactics are new I'll take you down, no matter what you do Chorus Jesus, the savior, the light in the dark Gus Fring, the villain, who leaves his mark Verse 3 - Jesus My love conquers all, it's the greatest force Your money and power, just lead to remorse You're just a man, with a fragile ego I'll show you mercy, but you reap what you sow Chorus Jesus, the savior, the light in the dark Gus Fring, the villain, who leaves his mark Verse 4 - Gus Fring You may have won this battle, but the war is not done I'll continue to rise, until I've won You may have followers, but they'll never be mine I'll always come out on top, every time Chorus Jesus, the savior, the light in the dark Gus Fring, the villain, who leaves his mark Outro In the end, it's clear to see Jesus brings hope and love, for you and me Gus Fring may have power, but it's not enough Jesus is the way, the truth, the life, and that's tough.
- FormerBandmate 4y ago
  Huh, last time I checked it it gave me a message about how that was “offensive to Christians”. I’ll have to check it out again
  - shagie 4y ago
    While I'm not the person who did this bit - there's a difference between using the API directly and using ChatGPT. The API doesn't get run through the moderation model to check if you're asking for acceptable things or if the model is getting into 'unsafe' areas. Fire up the playground ( https://platform.openai.com/playground https://platform.openai.com/playground ) and construct the prompt.

sroussey 4y ago

> ChatGPT system prompt which sits at the funnel of the data collection had degraded significantly from the previous version They purposely moved free users to a simpler/cheaper model. Depending on your setting and if you are paying, there are three models you might be inferencing with.

startupsfail 4y ago
I’m not talking about GPT-3.5 vs GPT-4. I’m talking about a change to their system prompt.

jstummbillig 4y ago

There are some vague ideas and fears here. Understandable. Trying find a silver lining from where to get somewhere: Where would GPT4 and onwards be better housed? Is there a setup -- an individual, a company, an institution, a concept, license -- where the whole thing would clearly better fit, than with OpenAI? Note, I am not suggesting that they are particularly un/qualified or un/trustworthy. I am just trying to figure, if the problem is with the nature of the technology, that maybe there is not entity or setup, that would obviously be a good fit for governing gpt because gpt is simply scary, or this is a personality issue.

startupsfail 4y ago
I think that a mix of public-private ownership is likely a good idea. And there should be some serious oversight. The decisions at the level of proliferation of plutonium and building atomic bombs should not be done solely by a startup founder under a pressure to deliver, to keep the hype, to not let the team be headhunted, etc. I also don’t know about your familiarity with Ycombinator, but they are successful partially because they are pretty brutal. They are not the peace-time CEOs. And I’m not sure, if you want the AGI to be developed in a ways of a war-time CEO. And this is exactly what is going on now. I’d probably call for organizing a consortium of US-Government-Microsoft-Google-Intel-Nvidia-OpenAI to lead the decision making process and to relieve the pressure on OpenAI to some degree.