Show HN: Interactive Graph by LLM (GPT-4o)

From a previous submission of this domain (May last year): <a href="https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=35798482">https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=35798482</a><p>&gt; After trying some queries, I found ChatGPT is good at finding public data, sometimes it may not be accurate, but overall, pretty good.<p>GPT-4o is no more reliable for this than GPT-4 or GPT-3.5 before it.<p>One of the many unintuitive things about LLMs is that they are terrible tools for looking up specific facts about the world - which is weird, because that&#x27;s traditionally one of the things computers have been best at.<p>If you want to build a working version of a system like this the trick would be to provide it access to tools for looking up data in citable, reliable data sources - things like the CIA World Factbook.<p>Then display the charts along with a link to that underlying reliable data.

Cute, but can I see sources on the data? I tried some query on some belgian statistics and it gave me just... very weird numbers that i do not trust, but zero way to check them.

This is pretty cool, it&#x27;s snappy, it seems to work, but there&#x27;s no way I trust it without a way to check the LLM&#x27;s work.

When prompted for<p>&#x27;daily unique visitors to openai.com by month since 2022 to 2024&#x27;<p>It gave a graph with 1) time axis decreasing left to right, 2) visitor numbers which can&#x27;t be real (near prefect linear trend) 3) points in the future, going out to Dec 31, 2024.

Hmmm this doesn&#x27;t look right? : <a href="https:&#x2F;&#x2F;columns.ai&#x2F;chatgpt&#x2F;AXoHKTTYYkeT3x" rel="nofollow">https:&#x2F;&#x2F;columns.ai&#x2F;chatgpt&#x2F;AXoHKTTYYkeT3x</a>

Looks about right <a href="https:&#x2F;&#x2F;columns.ai&#x2F;chatgpt&#x2F;Gajr7yJ9O5x57o" rel="nofollow">https:&#x2F;&#x2F;columns.ai&#x2F;chatgpt&#x2F;Gajr7yJ9O5x57o</a>

Would you mind explaining a little more on what is happening in the background?<p>- What is your core technical value add?<p>- Do you have the data sets in your own database and you are using OpenAI to query them?<p>- Looks like you have your own home built database?<p>- Are you using LLM Agents?<p>- I saw that you have Airtable integrations, are you able to do the same for any datasource including Airtable?

Nice one, though I can&#x27;t scroll the list horizontally thus can&#x27;t read columns beyond screen bounds (iOS 17 Safari, tried &quot;best ingredients that go with bacon&quot; query and the result was a table view)

I can&#x27;t believe there are no sources for the data. A few charts I tried seem off, but even in the cases where it looks like it might be right, trusting it without a source is a big no.

How is this in any way different than an RNG? The numbers are completely made up, does slapping an AI label on a RNG somehow make this unique or interesting?

I don&#x27;t why it&#x27;s flagged, it&#x27;s just a free service for people to play with.

GDP in Germany for the last 10 years<p>4k each year but the bar chart goes up<p>Sounds about right

I mean kudos, the UX here is so insanely good it feels like magic, but I don’t trust magic. Where’s the data coming from?

I get that free will means you <i>can</i> do evil. But, given the choice, why would you not only do it this baldly, but then go out of your way to show it off?

Where do the numbers come from?

I think what the world <i>really</i> needed right now was an LLM making up fake statistics, presented in a convincing high quality way.<p>There is just no way this could ever be a problem. Surely everybody knows that if presented with some data they need to do a deep dive into the actual sources instead of blindly trusting a graph.<p>I seriously want to know what was going on in the mind of the person who made this.

This might be a fun project, but it&#x27;s going to cause extreme misunderstanding for many people. People trust graphs, and users of this site are going to unwittingly spread falsehoods to for example r&#x2F;dataisbeautiful or even less reliable social media pages.<p>I&#x27;d urge you to take this site down as it will be net negative to the world.

Wow, this is fantastic!! Thanks! Currently writing my PhD and this can be really useful<p><a href="https:&#x2F;&#x2F;columns.ai&#x2F;chatgpt&#x2F;3Gk20NKW6D4vTt" rel="nofollow">https:&#x2F;&#x2F;columns.ai&#x2F;chatgpt&#x2F;3Gk20NKW6D4vTt</a>

The political bias in the results is astonishing and shows why these models should <i>not</i> be used for educational purposes. Just ask it a &quot;contentious&quot; question and notice it give biased and sometimes nonsensical results - &#x27;crime rate by political orientation&#x27; shows three &#x27;republican&#x27; states coming out on top with three &#x27;democratic&#x27; states filling out the bottom. When necessary it just seems to make up data to get the desired results, e.g. &#x27;murder rate by political orientation&#x27; talks about &#x27;Country A´ and &#x27;Country B&#x27; with, of course, &#x27;right&#x27; being far more murderous than &#x27;left&#x27;. It claims that &#x27;democratic New York&#x27; has the lowest crime rate. &#x27;IQ by political party&#x27; is another interesting example. Compare this to &#x27;trust in LLM output by political party&#x27; and maybe those &#x27;dumb republicans´ (who do not trust these models, together with the supposedly super-smart independents by the way) suddenly seem to be a lot smarter than all those &#x27;bright &quot;democrats&quot;&#x27;.<p>This site is a harbinger of the agitprop factories which are about to flood the &#x27;net due to the general availability of LLMs.