Google debuts more powerful “Ultra 1.0” AI model in rebranded “Gemini” chatbot

Confusing name shuffles aside, "Gemini Advanced" vies to catch up with ChatGPT-4.

See full article...

 

[AngryEssay]

Nah, that's a false dichotomy. I don't want Google to never cancel a product ever, I just want them to have a clear and coherent plan for supporting core products.
For example, I don't care if Google wants to experiment with a Zoom competitor, and I don't care if they try some new and bold things there and kill the project if things don't work out. But I don't want them to replace their official Android video call platform (twice!), especially if the new product is missing basic functionality present in the original.

It's a failure of management, of leadership as well and a testament to the ideology of "move fast and break things" taken to the extreme. If only others would learn, but as long as the VC dollars keep flowing into the next new hype cycle, nothing will truly change.

 

[SeeUnknown]

Bards were also known for embellished and imaginative storytelling, which is something AI already struggles with (via hallucinations). I'm not surprised they moved away from the name.

But "Bard" actually describes the technology actually does.

P.S. Thankfully Benj uses the better term "confabulation" rather than hallucination.

 

[Kenjitsuka]

and any data sent over a network and stored elsewhere is susceptible to hacking or interception.

Benj, please add "being sold" at the end.
For the rest, I was impressed you added some negatives!

 

[disarmyouwitha]

Benj still trying to make confabulation happen because he refuses to use hallucinate makes me smile every time. =]

 

[Sakul69]

Initially, it was Palm, then Bard, followed by Gemini, and now Ultra. I'm a bit lost on distinguishing the product's name from the LLM powering it. It's an avoidable name maze, similar to the other Google products. They struggle with settling on a name and sticking with it, portraying a lack of focus within the company.

 

[wondering22334455]

I asked Bard to make me a simple website with just one button. it said it can't do code.

I asked it for a recipe for flapjacks. it refused citing "sexual content" (obviously I wanted someone to jack my flaps!)

I asked it what the best browser was. it told me that information was blocked due to 'promotion of violent content'

I asked it when does it think Google will brick bard. It simply refused to respond.

 

[AnisonSinger]

They should call it Forbin

Colossus, surely?

 

[JoHBE]

But "Bard" actually describes the technology actually does.

P.S. Thankfully Benj uses the better term "confabulation" rather than hallucination.

The issue isn't so much the exact terminology, but the tendency to ONLY use the term when the output appears to be false. Because even when the output is interpreted as coreect, the process that lead to it was... exactly the same. The differenc wasn't caused by bugs or operating the system improperly. It always does what it does.The ENTIRE output is hallucination/confabulation, not just the stuff we tend to label as such after the fact.

 

[Throwback]

I gave Gemini a quick try and switched back to (Classic?) Assistant for 1 main reason. I have a couple of Assistant routines set up for when my alarm goes off (reads out weather, calendar and news) Gemini does not have that option in the alarm settings.

 

[entropy_wins]

Google is an ad company. They exist to show ads and nothing else.

If they had split Google into search ($CORPA) and ads ($CORPB) the following would be the mission.

$CORPA would scrape everything possible, scan any book possible, build the most powerful LLM models to allow searches to be made.

$CORPB would actively try and find every useful product combination to sell using voluntary information given by active purchasing users (remember. the wealthier you are , the less likely you will buy something with an ad....).

In this world $CORPA would not be ditching the cached copies of the web, because the "search" mission wants to keep all, where as the "ads" mission wants you to ONLY see ads (not even OLD ads!!!).

I will not be buying access to their stolen models (they are all stolen until they can prove otherwise), no matter what they call it.

 

[iamai]

"So whether we should accept the summaries, interpretations, and "reasoning" they present to us without challenge is the biggest question that faces the public as this tech rolls out."

There is no question that all output always must be checked for correctness. One encouraging thing I noticed about Gemini is that it has a built-in feature that "evaluates" the output (the Google search icon at the bottom) to help you "understand the results". Certain text in the output becomes clickable and it shows the source with a link. Easy access to the sources on which the output is based is very handy towards making it easy for the user to check outout. It should be built-in to ChatGPT, too. Copilot has similar functionality.

 

[benjedwards]

Benj still trying to make confabulation happen because he refuses to use hallucinate makes me smile every time. =]

True story: The AI definition of "confabulation" is in the Cambridge Dictionary now because my mom broke her hip.

While working on my story about confabulations in AI from Dec 2022-April 2023, my mom broke a hip and ended up in the hospital. At the time, I was looking for another term for "hallucination," since it was poorly defined at that point and unpopular in some quarters, and my mom was on painkillers that made her very loopy. One of her doctors said she was "confabulating" stories, and a light bulb went off in my head. It seemed to fit the creative gap-filling at play in LLMs. I put that term in my article about why AI models make things up, and it was reported upon by others and ended up prominently noted in Wikipedia's "Hallucination" entry, which many people used as a reference, and eventually the meaning gained enough steam to get into that particular dictionary.

Thanks, mom! She's OK now, by the way—fully recovered and no longer confabulating (for the most part).

 

[andygates]

Oh, this is why something called Gemini's mealy-mouthed cavilling over election results was trending.

Still shite.

 

[bugsbony]

Google Reader was killed more than 10 years ago, why isn’t anyone releasing a viable alternative? The answer is simple, it’s not a profitable product.

Because you never bothered to look ?
I've been using inoreader during all this time with no problem.

By the way, I will never forgive google for closing google reader, and yes they are shutting down many projects. But what the fuck does it have to do with AI ? AI is still in the research/experimental stage of AI, of course models changes, name changes... I'd be shocked if it didn't, google or not.
Same with the privacy policy, it will change, for now it's experimental so they can review everything.
I'm sure there is a forum somewhere for random bashing of <insert brand> because I don't like <insert brand>.

 

[dsrickard]

Here's a question:
Why is Google so interested in downgrading its assistant while also upgrading a human-computer-interaction tool of a high class?
Doesn't anyone at Google (or Amazon for that matter) recognize the inherent potential for integrating AI into their line of virtual assistants?
Everyone wants to monetize their products but nobody wants to actually use their head.
Isn't part of the goal of AI to create human-like conversational responses? Why is it that big tech is killing off their conversational interfaces?!

 

[Bigdoinks]

Talk about a no-privacy policy. Google employees reviewing an keeping Gemini chat transcripts is wild.

I can imagine warrants and subpoenas for Gemini chat transcripts are gonna become the new hotness in criminal and civil litigation.

And if a Google employee reviews a Gemini chat transcript that suggests someone is planning a murder or learning how to generate child porn, do they report it? Is there a duty to do so?

If Benj doesn't post an article in the next week, we'll know he is likely being interrogated by the FBI for asking a chatbot how to make a bomb.

 

[mstrcat]

Waiting for the name change to Ultron....

 

[matthieub]

Computer scientists still do not know exactly how they work in detail.

That's why we don't ask computer scientists but machine learning experts instead? They are not the same people