Comment by knaik94 | Hacker News Reader

Comment by knaik94 | Hacker News Reader

To give Google some credit, I think people would have been equally upset if they did release an AI that ended up being used maliciously or without proper filters. Google has a habit of shipping things while they are still in beta. Gmail kept the "beta" label after launch for five years. Not releasing unfinished AI products was a calculated move and I think it was an overall good decision, but also an expensive one. I am just worried about the longevity of it, since Google clearly doesn't care to prioritize it. The Bard ad mistake implies a deeper issue about resources being allocated incorrectly.

OpenAI clearly spent a lot time red teaming the chatbot to make sure the outputs were sanitized. Chatbot as a service is an easy monetization strategy from big companies. I wonder if ChatGPT servers will share resources with Bing search servers, I can't imagine they will, but if they do, the paid subscription will be a lot more valuable. I think users will react extremely negatively if Bing AI results embed affiliate links on every response possible. On the other hand, I can see myself clicking links of products ChatGPT recommends if I am intetionally asking about some sort of product, like if I ask it types/differences of climbing carabiners.

Research is very collaborative, so I understand why those at Google stay there. But it seems like MS and OpenAI are prioritizing having a clear pathway for novel AI work to reach users, and it won't be limited only to implementations that will help serve ads better.

Did Bard make a mistake?

        Q: "What new discoveries from JWST can I tell my 9 year old child?"
        
        A: "JWST took the very first pictures of a planet outside our solar system"

Source: NASA’s Webb Takes Its First-Ever Direct Image of Distant World - https://blogs.nasa.gov/webb/2022/09/01/nasas-webb-takes-its-...

So, it took the first picture of a (particular) exoplanet, which is not to be confused with the first picture (ever) of an exoplanet. There is a misunderstanding here.

I feel like there's some bigger mistake I missed regarding that controversy. People are aware of chatbots making things up. Microsoft had that poor AI Tay where users' adversarial prompts effectively courrupted its working dataset and it became racist. I know AI doesn't have feelings, and yet I still feel some sadness for how much it was attacked. Maybe because I liked the idea and wanted it to be successful.

Regardless of the validity, I think it's still fair to call it a mistake because of the ambiguity. Most people would misunderstand the answer to mean something it doesn't. But there was such a strong reaction in media about it, when I would have assumed most people wouldn't care. Google never promised an AI chatbot never got anything wrong, and no one I knew expected differently. It makes me think about how many other news stories, for things out of my field, are like the Bard situation. I'll just never know because I won't gain expertise in it.

I think the concern is that it reinforces impression that the Google announcement is rushed and half-baked in response to competitive pressures.

No. Another telescope had previously photographed this specific exoplanet. This information is in the linked article.

No interpretation of Bard's output here is accurate. It is wrong.

JWST took the first picture of LHS 475 B.

After ChatGPT's training cutoff of 2021. With the information the AI has in its dataset, it is incorrect. It's unaware of things that happened last month, and "put out wrong answers and wait to see if they come true" is not a business case OpenAI should rely on.

> So, it took the first picture of a (particular) exoplanet

c.f.

> Astronomers discovered the planet in 2017 using the SPHERE instrument on the European Southern Observatory’s Very Large Telescope in Chile and took images of it using short infrared wavelengths of light. Webb’s view, at longer infrared wavelengths, reveals new details that ground-based telescopes would not be able to detect because of the intrinsic infrared glow of Earth’s atmosphere.

The JWST news is that JWST took the first photo it has ever taken of an exoplanet.

Different planet.

JWST took the first picture of LHS 475 B, confirming it exists.

https://www.esa.int/Science_Exploration/Space_Science/Webb/W...

The source that Google cited states "The exoplanet in Webb’s image, called HIP 65426 b" which had been previously imaged.

Sure, JWST has also taken the first picture of a different exoplanet-- 2 months later. But JWST did not:

* Take the first picture of an exoplanet

* Take the first picture of the exoplanet in the cited source (instead, it took its first picture of any exoplanet).

I think this highlights a big limitation of LLMs. It doesn't "understand" the text. It's just piecing words together based on the probability. In this case, the key word was "its", which it ignored - but makes a huge difference to the meaning of the sentence.

Reply on news.ycombinator.com