It appears Microsoft Word is adding a ChatGPT function:
ChatGPT comes to Microsoft Word through Ghostwriter add-in | Windows Central.
Yep. It’s already rolled out into Microsoft Teams, too. It will do things like present summaries of meetings, action steps needed, and a whole lot more.
It’s rolling out into actual applications very quickly.
Yeah, this is what everyone is criticizing Google for. Palm is the best in the world (at least that’s been publicly debuted), but Google is very slow to put things in products. I’m sure Microsoft is going to gather a shit ton of feedback (reinforcement learning from human feedback “RLHF” is a big driver of chatbot success) from this, and in very relevant use cases too. As long as they have enough safeguard that it’s not suggesting putting the N-word into documents, this could be very good for Microsoft.
Microsoft is a bigger player here than many realize. Everyone is focused on what Google and Meta are doing with AI, but in the meantime the actual supercomputer infrastructure for training ChatGPT is hosted in Microsoft’s Azure cloud, Microsoft just put $10 billion into OpenAI (mostly in the form of credit for supercomputer time, which OpenAI is paying a fortune for), and they are not just rolling out enhanced office products, but Bing search is getting charged up with ChatGPT. This may finally be what Bing needs to take on Google.
If I were Google, I’d be very, very worried. In fact, they declared a ‘code red’ when ChatGPT came out, and even pulled back in their old executives for consultation.
Google has a big problem in that they get money for clicks, and therefore an AI that can give you an answer without serving up web pages kills their revenue model. And it’s about time Google got some competition: I find I use Google search less and less, and when I do I find it increasingly frustrating.
Google used to have an uncanny knack for serving up the best results right up front. Now, the first page is full of ads and SEO-enhanced garbage sites like Quora and Pinterist. The second page seems to be SEO and ad screwed as well, with the occasional good link mixed with stuff I don’t care about at all. Often I have to get to the third page before I start seeing the stuff I really wanted.
Google is compromised in that their revenue model is now at odds with user’s needs for efficient search. They probably held back their own AI for this reason - difficulty monetizing it. That’s how old legacy companies get destroyed by upstarts without their baggage.
Meh. Compute is a commodity. ChatGPT was only trained on Azure because a lot of Microsoft’s investment came in the form of credits. Google and Facebook both train plenty big models and Google has a large cloud business with arguably better hardware (although wide adoption has a benefit of its own).
Microsoft Research doesn’t get enough credit though leaving aside OpenAI. I work in object detection/computer vision and they’ve had a ton of great papers in the past year or so.
I don’t get this. Google’s business model is learning user preferences and selling that data to advertisers. If their search interface was an LLM they would still learn user preferences. Heck, that’s inherent in the LLM. You could still sell that behavioral surplus (as it’s referred to in the Age of Surveillance Capitalism) and serve ads on the interface page. I don’t think it changes their model at all. Huge change in approach though, and that’s risky, but the revenue model is the same.
Hey @Sam_Stone, this isn’t quite the death knell for Google you seem to think it is. The above article references “Ghostwriter”, a $15-25 third party plugin that does the equivalent of saving you from needing to copy-paste your ChatGPT output into Word.
While Microsoft HAS announced that they are looking at adding AI to Microsoft Teams and Outlook, it is not “already rolled out” by any means.
Right now, Google has an AI that is actually capable of finding reliable web sources for nearly any question imaginable. ChatGPT… does not. Google’s AI doesn’t format its responses conversationally, but it delivers correct information.
Sort of like the difference between a normal teenager and a gifted young child. They both may mature to the same level, but the gifted child has the potential to surpass the teen once it gets more real world experiences.
Regarding the nonsense it spits out about the number 2023, I suspected that since these models learn more about things mentioned often and don’t know about numbers or math, maybe chatGPT would be better with facts about a smaller number which likely is mentioned more often on the internet.
So I asked about interesting facts about 17
17 is a prime number. 17 is the sum of the first four prime numbers (2 + 3 + 5 + 7). In binary, 17 is represented as 10001. 17 is the 7th prime number and the only prime number that is equal to the sum of its digits raised to a power greater than 1 (1^3 + 7^3 = 17). 17 is the atomic number of chlorine, a chemical element.
Unlike the facts I got about 2023, which were all wrong, the above facts are all true.
1^3 + 7^3 = 21?
1^3 + 7^3 = 21?
I missed that 
(I had even checked the atomic number of chlorine, because I wasn’t sure)
Still much better than the facts about 2023
Google just signed a deal with Anthropic AI similar to Microsoft’s first deal with OpenAI; $300M of compute for 10% equity. Anthropic has a chat-bot called Claude that’s also very advanced. Not open to the public yet. Pretty interesting that Anthropic was founded by people that left OpenAI.
Today I learned ChatGPT can answer Jeopardy questions. It’s really good at it in my limited testing. It even gives the answer in the form of a question as long as you tell it that’s what you want.
OpenAI has started rolling out their paid ChatGPT plus service. I already have the link available to upgrade on my screen, but YMMV depending on where you live and whether or not you put yourself on the waitlist.
Free Plan: (what everybody starts with)
Available when demand is low
Standard response speed
Regular model updatesChatGPT Plus $20/mo:
Available even when demand is high
Faster response speed
Priority access to new features
I’ve decided those extra features are not worth it to me, as long as I can still find some time when the free tier is not too overloaded. If the paid service offers an option to take down those safety guardrails that tend to result in glurge & pablum in the stories it generates I might go for it.
And now that “Upgrade Plan” button is highlighted in yellow on my screen. I guess they really want to make sure eligible users see it. I might not have noticed it at first except that it was taking up extra real estate on my screen, stealing area from the chat history list.
Aaand today is why Google has been slow to ship AI product. A factual error from their chatbot and they lost $100B in market cap.
Link to something?
Word of advice, whenever an article uses the word “after” in the title, you’re being lied to (especially in financial news). For example:
Google shares plummet after chatbot ‘Bard’ gives bad answer
There is never any evidence that the two events are linked. More often than not, they didn’t even get the temporal order correct. It’s just as likely related to the recent layoff news or a million other things.
So “after the Fed raises rates” or “after worse than expected earnings?”