The key question here isn’t so much whether GPT-4 beats the actual human decks as much as what it had to fabricate to do so. The humans are probably constrained by things like “reality” and “what their business has done in it” while GPT-4 could make up anything it wanted. A fair comparison would be to humans given the same prompt and told to invent whatever facts they wanted.
> 1 in 5 investors and business owners pitched by GPT-4 would invest $10,000 or more.<p>This is a completely useless metric and also a common pitfall. It's been shown many times that what people say they would invest/pay for something is completely unrelated to what they actually do when push comes to shove.
I think this is actually a good thing.<p>It seems there are a lot of people who have no talent but being able to hype up ideas. Often they have overshadowed the actual people who come up with and implement ideas.<p>GPT-4 and LLMs in general democratizing generating BS may actually shift the advantage more to the actual doers.<p>For example, There are probably on HN brilliant developers who have written at the core of really groundbreaking services and applications but will never be able to get consideration because they just don’t have the talent for making slick decks. Them being able to use GPT-4 to generate pitch decks would allow them to get their ideas in front of a lot more people.
In my experience having pitched to VC's it's not that there is a perfect deck but many really bad ones.<p>I.e. there is diminishing returns on a "perfect" pitch deck. You just want to be in the category of not sucking and the decks level of perfection is no longer important.<p>Today I would even go as long as to say the pitch deck is much less important than a demo.<p>You can in theory pitch in an email, the rest is just to make everyone feel good about giving and receiving the money.
I find it weird to compare GPT-4 to not expert humans (which this did - they seemed to have just sampled pitch decks for funded companies). GPT-4 generally will beat non-experts.<p>What you want to see is a comparison to say pitch deck consultants - it's also more useful to evaluate whether AI or a human should help you build one.<p>This does suggest at minimum though that if you are a founder that GPT-4 will likely provide positive value in helping you relative to no one.
This is how we lose control. We won't have to wait until the AIs "wake up" and take over the nukes.<p>Humans are very susceptible to bulls** which might be the most powerful force in the world. AI bulls** generation is already superhuman and continues to accelerate. The only way to counter it will be with your own AI bulls** generator going head to head with the competition.<p>In the coming years, it will go faster and faster until it is dozens and then hundreds of times faster at generating high-level bulls** than humans. Due to their superior powers of persuasion, these AIs gather more and more power. Civilization ends in a hyperspeed barrage of superbulls** fighting superbulls**.<p>It's already started. <a href="https://m.twitch.tv/trumporbiden2024" rel="nofollow noreferrer">https://m.twitch.tv/trumporbiden2024</a>
But does GPT-4 outperform a skilled human BSer.<p>We sometimes see BSers of various degrees in business. Four thoughts:<p>1. A lot of things are seem easier for someone who knows what would be an advantageous thing to say in the moment, to that audience, and just says it, unconstrained by truth.<p>2. Some people are so good at BSing and complementary skills that they're incredibly slick at presentations. You might want to have them as a frontperson, until you consider...<p>3. Chickens tend to come home to roost. BSing is a scam tactic, and people tend to get burned by that.<p>4. Often the BSer themself seems immune to adverse repercussions.<p>GPT-4 is Robo-BSer.
I would be interested to know how this was done.<p>Unless they have access to Code Interpreter then gpt4 won’t generate graphs and images.<p>Did they have it generate Python that made the graphs in matplotlib or something ? What did the prompts look like?
I bet I could improve your prompts and make the pitch decks even better. Let me know if you would like a prompt audit. I also have a detailed prompt for content analysis and scoring. The reason being is because I do tests like this as well to compare GPT-4 basic prompts to prompts enhanced by the SmartGPT AI tool I created.
Dunking on VC is a useless takeaway.<p>Ignoring snark, VC's and founders giving pitch decks are above-average educated intellectual demographics. This shows how much more convincing current LLM are than the majority of the population. Social media can have regular peoples' voices drowned out by a flood of more persuasive bots. Instead of actually talking issues out amongst each other, even on HN, it could be a deluge of output from "Give me the most convincing argument why they stole the election from Trump."<p>Now imagine what will happen with state of the art LLM in a few years...