Until next year I guess.

bard already can get captions and understand videos @@1fattyfatman 💀

That’s a pretty meta question

Gemini respond , show me the duck and I’ll tell you the story 🤣🤣

ah. And it also poses an existential threat to humanity. But that would have made it 94 seconds, so it had to be omitted.

I'm genuanly frigthened by theese claims! Is it dishonest marketing, or an incredible revoloutinary technologial leap?

Thanks for the insight

@@allyouneed247 it is impressive, but too me it looks like performance improvements rather than a new paradigm or technology. It's getting good, yet the complexity of the problems it handles are not of a different class from what we have seen before, just a higher level of difficulty. I have only seen a handful of examples today, maybe they will show more fundamentally different types of problem solving later, but I am not really expecting it.

couldve been in the training data

@@EskiMoThor going from 1000 to 3200 elo is a paradigm shift, i’m 1000 elo in chess, magnus carlsen is 3200

So we are officially in the Matrix and Terminator films 🤔

​@@deneilwilliams4380No, because those are works of fiction, and we live in a reality where stupid people like you think movies are reality.

​@@deneilwilliams4380🎉x

​@@SteveWood-gk5ptAI has been programming itself using Neural Architecture Search since as early as 2012 buddy.

0:07

kzfaq.info/get/bejne/oLxhqc5525reoWw.html

kzfaq.info/get/bejne/oLxhqc5525reoWw.html

Yeah after all we are teaching it. Think about it.

they posted all the benchmarks

Do you remember the Pentium 4 with the shameless NetBurst architecture with the GHz that were not what they seemed? This was back in the day where Intel made the fateful decision to compete on brochure allure, rather than something deeper under the hood. It did finally catch up to them, but it was a close run business. AMD very nearly did not survive. Intel's current market cap is $180 billion. AMD's market cap is $280 billion. Nvidia's market cap is $1.8 trillion. Who's laughing now? What was that famous lyric from long ago? _Benchmarks don't mean a thing, if they don't foreground the bling._ For a long time people lapped it up, being the clueless sheep they generally are. Prescott from 2004 was a low point in human history. "Prescott's inability to achieve greater clock speeds was attributed to the very high power consumption and heat output of the processor." The Western world built entire additional power plants to feed this horrific aberration in effective power/performance ratio, all in the service of bragging about a GHz figure that didn't deliver what people assumed after a long run where clock frequency was a reliable proxy. The truth of the matter is that we will never have a proxy like clock frequency again, as it progressed from the late 1970s through to the early 2000s. More recently, a certain politician showed up promising to return America to the glory days of the 1950s. The truth of the matter is that 1945 to 1975 in America were the three most glorious decades that any country has ever known in all of recorded history. Every major economic adversary was either vanquished outright or digging itself out from the rubble and financial ruin of WWII. Except for Stalin with the H-bomb, there was barely a cloud on the horizon for ascendant America. Out in the suburbs, there were two huge gas guzzlers in every driveway-his and hers. Los Angeles was almost a mythical city. No one had ever seen the whole thing, because of the thickness of the sky. PC Magazine was famous back in the day for its benchmark culture. In particular, with the new wave of accelerated graphics cards, everyone poured over pages and pages of benchmark charts. The PC Magazine benchmark suite was loser city, however. They tested sprite transfer rate by drawing the same small sprite over top of itself 10,000 times in a row. Pretty soon some of the vendors noticed that they really didn't have perform the last 9,999 idempotent repeats. One of those vendors was ATI (since auctioned off to AMD). All you had to do was faithfully copy the first one, and then you could treat the next 9,999 iterations as no-ops. Sure made your bar bulge out from the heap in the graphical aftermath. I'm one of these people who reveled in figures of merit, when I could get some. In many ways memory bandwidth was a better metric than the usual CPU benchmarks of some silly Photoshop filter. The original Pentium Pro trounced most of its RISC competitors on memory bandwidth agility. It rarely won any single benchmark, but you could load a single machine up with your entire back office and it never slowed down. Meanwhile, the RISC systems were a lot like American automobiles from the 1950s: fastest cars on the road in a straight line. RISC continued to brag about the elegance of their core architecture, because an elegant core architecture could be designed by a small team of designers. This made it possible to pretend to compete with a behemoth like Intel. Meanwhile Intel assigned hundreds of engineers to the cache and memory hierarchy, which is far more difficult and expensive to perfect than the rest of the CPU core. In SMP systems you have to do a lot of cache snooping to decide if you own the memory or not. When you don't, it generates highly unpredictable memory latency. Unpredictable bottlenecks are everywhere. We talk about embarrassingly parallel, but we should have talked a lot more about embarrassingly predictable. Photoshop filters were embarrassing on both fronts. They tended not to argue much about which core owned which cache lines. Hence every RISC benchmark ever published by Steve Jobs included a Photoshop filter. But as soon as you got twenty asynchronous daemon processes running on your back office server, that never happened again. The figure of merit that mattered to 90% of the hard-blinking population should have been cache agility, but it never was. Instead we got benchmarks that fixated on peak parallel dispatch rate. In special cases PowerPC could dispatch four operations on the same clock cycle. In the normal condition of pipeline bubbles caused by unpredictable memory latency, the PowerPC sat there dispatching zero. The Pentium Pro had a peak dispatch rate of three operations on the same cycle. If one dependency chain stalled, it would continue bursting along at three operations per cycle for another five to ten clock cycles. This was often long enough for an agile cache hierarchy to clear the bottleneck. That was the glory of out-of-order execution: it made the agility of your cache hierarchy even more apparent. Steve Jobs worked overtime to keep this particular story off the front page. Sorry to report that even back in the glory days of reliable proxy benchmarks, the reliability and suitability of such was greatly oversold, _especially_ by one of the greatest oversellers who ever lived.

This means Google is worth investing in the stock market right?

@@lucasrateau7580 Yes!

Next year, we'll definitely know when because Sam Altman is going to announce GPT-5 at the same time! 😂

Never

@@wyqtor GPT-5 based on Q*

Literally: in ten days. These people are idiots, don't trust KZfaq commenters. Their site tells you.

never

Optimus will be running on Grok😅

I don’t expect it to be open-source or locally available, if this is as good as they say it is they are going to try and profit off of it by making it proprietary

​@@rebucato3142Even if it was open source, I doubt any ordinary person can run the ultra model. Nano will be in pixel phones though. Pro might be the ChatGPT equivalent.

Later this year, no, Vertex

0:49 burnol for SUV peeps

it is now if you are still wondering

Take a deep breath and think step by step, also i have no fingers...

There are two "e"s in ketchup. ChatGTP 3.5

Gemini is a whole system, Bard is just an Ai assistant

@@michalotron actually I know, it just too many brand names in Google ecosystem.

There's a difference between an LLM, a base model, a fine tuned model, and an agent, and you people are stuffing them all under one, two letter title.

Can't wait for you to get prompting using Bard with Gemini Pro ✨

​@@Google ,C.

Bard is powered by this. Until now, it is powered by PaLM2

thank you @@ravitejaknts

Why?

Probably some custom ML framework.

teaser as far as i can tell...

because even so gemini is projected for resolve a asks or tasks, he yet not have conscience

Yes! And launch date? Google, plz

And my data would be safe and untouched without it. :P

No you wouldn't. Whatever Google did to happen to be there when your life was saved, had Google never existed, something else would've.

They added it to bard yesterday 😎

Bard exists....

We sure think so! 🌟

no?

@@humand0969 I never thought a person would be able to come up with this comment

🤡

What open? OpenAI's ChatGPT models are not open source.

From today onwards 🎉

@@dragondevelopers641in fact, it started yesterday. :)

They aren't open source.

@@xE92vD thx where can i download this opensource (

P

OduJFiqaydlaieapâpqpaepqoowO eş aopalpspp EA ilwaala watt a 0isjp

Erotika aks

Competition is the whole point. Google wouldn't develop gemini if it's wasn't OpenAI ChatGPT. OpenAI wouldn't speedup the GPT-4V multimodal if it's wasn't Gemini

@@zeon3123 Yeah you're right. I'm just commenting on the way they present it. Like it's something groundbreaking.

That is a fallacy very well known in economics.You can search it up. Populations tend to believe that jobs will disappear after every major tech revolotion. But the truth is that it may indeed change the labor conditions, but will create new jobs in other fields

Exactly!!👍

Just use a VPN😂.... I got access to bard thanks to VPN... It's not that hard ya know

It's 2023 and if you don't know how to use a vee peen then you literally don't deserve to be online.

Stoked for you to take Bard with Gemini Pro for a spin 😊

​@@GoogleNice

Bard

And we couldn't be more excited about it!

Only two of those are LLMs. Not all AI are the same thing buddy