Google Begins Testing Its Own ChatGPT-Style AI (gizmodo.com) 19
Google is rushing to release its own artificial intelligence products in the wake of OpenAI's ChatGPT. From a report: The search engine pioneer is working hard and fast on a "code red" effort to respond to ChatGPT with a large language chatbot and testing new ways to incorporate that AI-powered bot into search, according to a report from CNBC. The new report backs up earlier news from the New York Times and elsewhere, which outlined a rapid re-alignment in Google's priorities in direct response to the rise of ChatGPT. CEO Sundar Pichai reportedly re-assigned employees and "upended" meetings to boost the amount of resources going towards the company's AI development.
CNBC's Tuesday account offers further details. Google's new chatbot, reportedly named "Apprentice Bard," is based on the company's pre-existing LaMDA (Language Model for Dialogue Applications) technology. The application looks and functions similarly to ChatGPT: Users input a question in natural language and receive a generated text response as an answer. But Apprentice Bard seemingly has a couple of important skills beyond what ChatGPT can do. For one, it can draw on recent events and information, according to CNBC, unlike ChatGPT which is limited to online information from before 2021. And it may be better at achieving that elusive AI accuracy. For instance, LaMDA correctly responded to a math riddle that ChatGPT failed to grasp, as recorded in company documents viewed by CNBC.
CNBC's Tuesday account offers further details. Google's new chatbot, reportedly named "Apprentice Bard," is based on the company's pre-existing LaMDA (Language Model for Dialogue Applications) technology. The application looks and functions similarly to ChatGPT: Users input a question in natural language and receive a generated text response as an answer. But Apprentice Bard seemingly has a couple of important skills beyond what ChatGPT can do. For one, it can draw on recent events and information, according to CNBC, unlike ChatGPT which is limited to online information from before 2021. And it may be better at achieving that elusive AI accuracy. For instance, LaMDA correctly responded to a math riddle that ChatGPT failed to grasp, as recorded in company documents viewed by CNBC.
3rd time's the charm (Score:1)
Three ChatGPT [slashdot.org] stories [slashdot.org] nearly back-to-back. Is there some kind of great investment opportunity that I'm missing out on *rolls eyes*, or are the editors just really excited about AI today?
Re: (Score:3)
Re:3rd time's the charm (Score:5, Insightful)
This might be the most significant battle among internet giants in a very long time.
Re: (Score:1)
Call me a luddite if you must, I just find it hard to get excited about something that is being developed to put humans out of work. Perhaps it wouldn't be so bad if we had a Star Trek post-scarcity society, but being unemployed in the real world means poverty and no health care (yeah, it's an American perspective).
Most prior innovations which caused job loss as a side effect produced tangible improvements to human standards of living. IMHO, I'm tired of talking to machines - please, businesses, bring bac
Re: 3rd time's the charm (Score:1)
That's why so much coverage. It's happening fast and most have no idea how it works. The transparency the internet was supposed to have as promised is actually causing some pain? A kind of stinging ignorance.....
Which part of the company (Score:2)
I wonder which company that will end up with once Google gets broken into pieces in the antitrust case.
Mythical Man Month (Score:2)
Re: (Score:2)
Training large models requires a lot of human oversight and review of literally millions/billions of atomic, non-interrelated "things" (depends on what is being trained for).
This lends itself perfectly to "throwing bodies at the problem", because bodies are actually the solution - parallel works in this situation.
Building software doesn't work like that. Tasks are still atomic (distilled down far enough) but are inter-related and not capable of parallelization.
The search engine pioneer.... (Score:1)
Yep lets ignore all the search engines prior to google's existence...
Yep they never existed...
Awesome journalism as usual by Gizmondo...
Re: The search engine pioneer.... (Score:1)
Get your .Net books out of the basement, just saying.
Use ChatGPT to filter/rank results? (Score:5, Interesting)
Everyone talks about using ChatGPT as mechanism for generating results to search queries, e.g. you type in a factual question "when was the mona lisa painted?" and google/bing will provide a ChatGPT answer above its normal results.
But I wonder if there's an entirely different way of using ChatGPT for web-searches? Could ChatGPT be used as an input for filtering/ranking/categorizing the existing results?
1. Could ChatGPT assess how well the page answered the question?
2. Could ChatGPT assess whether the page was itself machine-generated (and hence downrank the page)?
3. I know that ChatGPT is able to explain the chief opposing views on a topic; could it be used to judge whether the page delivers something in line with one of the main opposing views, or a fringe theory?
4. What if ChatGPT were asked to deliver a summary of each indexed page, and then when you search for a term it would prioritize results where the term appeared in the summary?
5. What if ChatGPT is used internally to answer your query, or even just figure out what it means in cases where people write barely coherent queries, and then google/bing does a web-search for pages that are close to ChatGPT's answer as well as those that are close to your query?
6. Could ChatGPT internally generate a summary of each result, and group together results that are mostly identical or repeat the same information?
Why?
So often, when I search on google, there are a few pages of almost identical results written by journalists who put minimal effort into generating filler, or copying what a press release had said without applying any insight or analysis. I'm happy to read one or two such results, but wading through all of them looking for a nugget of useful analysis is frustrating. It'd be nice if ChatGPT or similar could help avoid this.
Whenever I look for help with an iPhone or Mac issue, there are loads of pages that tell you steps to follow, but very few pages that analyze those steps or explain what they're doing under the hood or why they should work. As a tech nerd, I want to know! I think that something like ChatGPT would be able to recognize whether a page contains analysis or just repeating steps.
Often I find that a simple query like "is it safe to replace a light fixture without turning off the breaker" gets me pages that have 20 paragraphs of filler "what is a breaker? what is a light? what is a fixture?" before they get to a two-sentence answer. I assume they do this because SEO rewards it. I think that ChatGPT-level "understanding" of text would be sufficient to downrank this kind of useless filler.
When I do searches about video games, a lot of the content is written by games journalists who don't have any original information to report and are just writing out in several paragraphs of long-hand what should have been seven words of bullet points. I don't know what the solution might be, but I'd love to see something. change.
Re: (Score:2)
You allude to ranking, filtering, classification, summarization, elaboration, entity extraction... all that is in the works.
There are still bugs to work out (Score:2)
Another Gizmodo Article questions the AI's usefulness in its present incarnation.
TLDR:
There’s a lot of opinions flying around about the pros and cons of the future of ChatGPT, but Arvind Narayanan—a computer science professor from Princeton University—thinks the hype is still overblown at this point.
In an interview with The Markup last week, Narayanan expanded on his previous claim that ChatGPT is a “bullshit generator” in his newsletter on AI snake oil. Narayanan argues that since ChatGPT isn’t trained to produced true text, it’s trained to produce plausible text—accuracy is a side effect of its goal of being persuasive. This means that ChatGPT’s output might usually sound correct but will eventually fail spectacularly and produce blatant misinformation, which Narayanan says isn’t a good fit for something like education, or journalism.
Search is going to change (Score:3)
Smart AI search is expensive - very expensive in terms of CPU & resources. OpenAI is already charging $20/mo for ChatGPT+. Microsoft and Google won't be able to give this away to the public as it has search. The ads models of today do not fit well in the new models.
The Search market is likely to be upended in ways that some will celebrate and others will curse.
Quality search is going to become a subscription service and just about every white collar job is going to pay for it because you won't be able to compete without it. As an upside, ads will be going away, or at perhaps going away for premier tier access. Lower tier access may still see a few ads, but they will be clearly separate from the results.
Most blue collar workers will likely remain content with old-style free but ad-supported search.
The digital divide is about to explode.
Re:Search is going to change (Score:4, Funny)
...or they'll just insert ads into content... e.g.: ``write a story similar to little red riding hood, but in a way that makes me want to drink a diet coke.''
ChatGPT:
Once upon a time, in a bustling city, there lived a young woman named Cherry. She was known for her bright red hoodie and her love for diet coke. One day, she decided to visit her grandma who lived on the outskirts of the city.
As she walked down the busy street, sipping her diet coke, she felt a sudden thirst for another one. She stopped at a vending machine and got herself another can of ice-cold diet coke. As she continued her journey, she met a sly wolf who tried to trick her into telling him where her grandma lived.
But Cherry was smart and didn't fall for his tricks. Instead, she offered him a sip of her diet coke and told him that she was on her way to see her grandma. The wolf was taken aback by her kindness and decided to join her on her journey.
Together, they walked through the lush forest, chatting and sipping on their diet cokes. As they reached the cottage, the wolf got the shock of his life when he saw that Cherry's grandma was not the frail old lady that he had expected, but a fierce and independent woman who was a huge fan of diet coke too.
The three of them spent the day chatting and drinking diet coke, and the wolf even helped them with the chores. As the sun began to set, Cherry and her grandma offered the wolf a bed for the night and even gave him a can of diet coke before he drifted off to sleep.
The next morning, the wolf left, a changed man, with a newfound love for diet coke and the kindness of people like Cherry and her grandma. And Cherry and her grandma lived happily ever after, always keeping a steady supply of diet coke in their fridge.
The end.
Feeling thirsty yet? Grab yourself a diet coke!
Google assistant (Score:2)
I hope they make something better and replace google assistant with it. Google assistant can't find things on my shopping list because it starts with a capital letter. Manually change something to lower case and suddenly it can find it.
ChatGPT Therapy (Score:2)
I just asked ChatGPT a question about some troubles with a lunatic manager and it's response was a warm blanket compared to my folks.
Whether that's an endorsement of it or a criticism of my parents or both I really don't want to touch
ps (it's both)