Bing chat is a rushed product. I can’t believe they let it use the company brand so easily. Didn’t anyone in Microsoft know how the public used ChatGPT and that Meta’s Galactica LLM before they got filtered and shut down respectively?
This headline wouldn’t have happened if they just let it be Sydney.
Microsoft, in general, doesn't seem to think before releasing things like this into the wild. Have we already forgotten about "Tay" the racist chat bot by them? [0]
The scary thing here is that Bing is exposing why you shouldn't use Microsoft products. Their AI is weaponizing the things folks have been saying are benign data that's collected from search history. If you're viewing this from a Microsoft OS you may want to rethink the platform you're using may use it's telemetry it's collecting about you against you in the future if (and when) it doesn't agree with you.
I don't treat it like a person, but rather as ChatGPT that has access to Bing's search index. For factual queries ("who invented X? at what company?") it's more reliable than ChatGPT and saves me quite a bit of time. Similarly for content aggregation tasks - I'm *scared* to click on a lot of "Top 10 X", "The best Y in Z" type pages because they're SEO and advert mined. Bing does the initial aggregation for me, and if interesting, I ask it to tell me more about that resource or visit the page myself.
There will always be a segment of people who troll on the internet - that doesn't detract from the immense productivity tool it can be when used correctly.
> it's more reliable than ChatGPT and saves me quite a bit of time.
What evidence is there that the results you get now are significantly faster or better than the non-AI results you got before? Particularly for queries of the form in your example.. I really wasn't having much difficulty getting those answers before these products existed.
- the low-hanging fruits (where there is a Wikipedia page or similar for X, and both Google and Bing do a good job of mining these)
- the tougher nuts ("who was the UK prime minister when the wright airplane first flew" - Google and regular Bing fail at this, but Bing chat correctly brings up Arthur Balfour). This was just an example I made up to try - but the ability to connect more dots than plain old search, which is hard to explain but you get a sense of the capability as you use ChatGPT/Bingchat - helps a lot.
The search LLMs are good at synthesizing answers that don’t appear anywhere on the net. But they also hallucinate answers often. So to get reliable results, one needs to fact-check them. Otherwise, the risk of being misled is high. The fact-checking isn’t much faster than just looking up different bits of information and synthesizing the answer oneself.
There are cases where LLMs make life a lot easier for people, but I am not convinced about whether search can be made easier by the way Sydney and Bard do it.
If they suggested alternative search queries and summarized websites for their search result excerpts, the LLMs would speed up search a lot. They could also synthesize some content quality metrics for each search result and highlight ones with biased reasoning, political influences, SEO games, and so on.
I too am optimistic about the tech. And I would even consider myself an early adopter of convolutional NNs and these LLM products in my commercial work.
But we have to state the obvious - Bing chat is not a good substitute for web search now. It is simply too unreliable. You.com has a better implementation of a search chat LLM imho because it quotes web more verbatim and uses more reliable sources. It also doesn’t simulate going on emotional tangents.
Sydney needed more time to mature as a product before Microsoft slapped “Bing” on it. It may still mature, but Microsoft took a big reputation hit for rushing this to market.
Oh I agree with that. LLM doesn't give you better search - better search {index, ranking, SEO-tolerance, ...} gives you better search.
To me, Bing chat - although imperfect - augments search positively to fill my specific needs.
0) I don't see half a page of ads before the first real result, disguised to look like real results.
1) The interface is clean and noise-free : it takes away a huge context switching load I incur when going into individual results (which for most searches these days, is the top 3-5). I just want the content summarized, with no ads, in a form my brain is used to.
2) I can ask follow-up questions with context, again without ever leaving the interface. Otherwise, the follow up question's answer is often on another website.
3) I can ask more creative questions, which is not really a 'search' feature. Something like 'write a snippet of code'. You can try "unique_ptr in rust - show examples too" and it gives me a passable and concise answer. It presents two options, but to get what I exactly want, I can ask "how to use Box?" as a follow on.
4) It's vastly better at 'connect the dots' queries - see my other comment please.
One underrated feature is the 'next query' suggestions - I can use a single click instead of typing out 'one more example' or see more subtle examples by clicking 'how can I use Box for recursive types?'.
This headline wouldn’t have happened if they just let it be Sydney.