Which of these 10 things will GPT-5 do? (@flowersslop)

1kṀ1297

Dec 31

49%

1. @flowersslop must prefer its output for any prompt at least 80% of the time vs GPT-4o. No exceptions.

25%

2. It must be able to create an acceptable Doodle Jump clone, end to end, no bugs or flaws, and it must look genuinely beautiful — at least 3 out of 5 tries.

36%

3. Image generation v2 must look at least as good as Midjourney v7, while being at least as intelligent as 4o imagegen. +1 Bonus if it’s uncensored.

40%

4. Avm must be way better. Sesame level minimum.

31%

5. Some kind of better personalization that isn’t just sloppy memory or static custom instructions. Something that actually feels personal and cool, like midjourney customization.

19%

6. Better or overhauled UI. Not just current ChatGPT with GPT-5 slapped in. @flowersslop wants it to feel new.

34%

7. Needs a new gimmick that’s cool or fun or useful. Something fresh. Like avatars, proactive texting, or anything novel that no other LLM provider is doing yet.

26%

8. Normal people who don’t care about AI must feel that GPT-5 is a lot better. @flowersslop wants to see demand explode.

31%

9. All features that make GPT-5 unique or cool must be available instantly via ChatGPT Plus. Only exception tolerated is a GPT-5 Pro

18%

10. It must not make any dumb, viral mistakes in week one. No 9.9–9.11, no “how many Rs in strawberry” type errors. Nothing memeably stupid.

See @Flowers ’s tweet:

If 0 to 3 of these are met, it’s terrible and I’ll say so publicly. AI was a bubble. AGI timelines get pushed back 3+ years.
If 4 to 6 things are met, it’s a slight disappointment, but im still happy, they are still on track.
If 7 or 8 things are met, then im really happy. It actually deserves to be called GPT-5. Solid work.
If 9 or 10 things are met, I’m really, really excited. This means AGI could actually be close. OpenAI proves it’s still number one without a doubt.

Resolves to Flowers’ judgement of which of these happened vs didnt happen when GPT-5 comes out. If I am not able to get access to Flowers’ judgement on these, they may resolve N/A or I may resolve them anyway if they are particularly unambiguous.

This question is managed and resolved by Manifold.

#AI

#OpenAI

#Technical AI Timelines

#GPT-5 Speculation

#ChatGPT

Get

1,000

to start trading!

3 Comments

22 Holders

78 Trades

Sort by:

bought Ṁ5 YES

I'm not sure how doodle jump clone is supposed to be beautiful. Is the model expected to generate both code and assets? Does it have to do it in a single prompt?

bought Ṁ40 NO

@ProjectVictory mb, I'll let flowers be the judge of that. if that's not acceptable feel free not to bet but I don't want to force my own interpretation of it on them

@Bayesian I won't be betting on that one since they don't set a compute budget or talk about turns. If they're expecting it zero shot from a prompt, it seems really implausible in part because recent OpenAI models have stopped and asked for clearer instructions before proceeding at a much higher frequency. If it is allowed to saturate the context then it seems very possible

Related questions

Related questions