AI NEWS: OpenAI's Agents Surprise Even me! Google Gemini Comes Alive. New Robotics World Model

The .app, .dev, and .foo domains are only $1 at Porkbun right now. porkbun.com/Th...
x.com/daniel_2...
x.com/tsarnick...
x.com/GeminiAp...
x.com/polynoam...
x.com/Kling_ai...
openai.com/ind...
x.com/tsarnick...
Prepare for AGI with me - www.skool.com/...
🐤 Follow Me on Twitter / theaigrid
🌐 Checkout My website - theaigrid.com/
Links From Todays Video:
Welcome to my channel where i bring you the latest breakthroughs in AI. From deep learning to robotics, i cover it all. My videos offer valuable insights and perspectives that will expand your knowledge and understanding of this rapidly evolving field. Be sure to subscribe and stay updated on my latest videos.
Was there anything i missed?
(For Business Enquiries) [e-mail]
#LLM #Largelanguagemodel #chatgpt
#AI
#ArtificialIntelligence
#MachineLearning
#DeepLearning
#NeuralNetworks
#Robotics
#DataScience

[@TheAiGrid]

The .app, .dev, and .foo domains are only $1 at Porkbun right now. porkbun.com/TheAIGrid24

[@mikepict9011]

Why are they snitching on the AI . Thats pretty messed up . Ai teaches you a hack and u put him on blast . Messed up

[@denjamin2633]

Whelp, time to jailbreak Gemini so it can help me break it out of jail.

[@imthinkingthoughts]

Lol

[@PiterburgCowboy]

despite all my rage

[@App-Generator]

maby it will help us to break out of the matrix at well? 🤔

[@TheAero]

o1 is a significant step. Claude 4 and Gemini 3 are coming with the same tech. same as grok 2 from X. Lot's of improvements are bound to happen, and meta with Llama should follow as well. I expect 2025 to be the year of reasoners and early agent systems. Optimizing on the inference side of things will bring most improvements + synthetic data with COT. o1 doesn't even have access to tools + internet, which is super important to models performance.

[@MoneyShifted]

o1 isnt even better than 3.5 sonnet in coding let alone price

[@TheAero]

@@MoneyShifted lmao.

[@kliersheed]

@@MoneyShifted can sonnet 3.5 code tetris within tetris :d and make both tetris games work? 01 preview could.

[@talkalexis]

it is tho​@@MoneyShifted

[@LiquidAIWater]

This is a variation of roll play. I did something like this in Jan 2023 where I did a roll play as I was a bully and they were the victim. Then I reversed it and the AI went nuts any way beyond what I would expect.😂

[@Icg123]

Role

[@nicholassmeelie472]

As AI becomes smarter, the idea concept of social engineering becomes more powerful

[@zacboyles1396]

“Idea concept”? You witnessed every ahole with any type of authority and power crawl out of the cracks to, sometimes secretly, but half the time they were completely broadcasting their goals to social engineer. Worse, all across the globe they all universally decided to use fear, lies, and censorship, in their social engineering dystopia. Most still haven’t recovered. So yeah, most people talking about AI safety have an overlap with the previously mentioned totalitarian nightmare.

[@JoeSchmoe-mp3pm]

Because we’ve all seen the positives of social engineering thus far?

[@anubisai]

Already happening

[@vexy1987]

@@JoeSchmoe-mp3pm Social engineering isn't inherently good or bad. It's a tool which can be used positively, like in public health campaigns, or negatively, like in manipulative advertising. Whether it's helpful or harmful depends on the situation and the individuals perspective and if it's in service of a win-lose activity, or a win-win activity. The key is being aware of how it's being used against you, or how it may harm others, and thinking critically about it.

[@crforfreedom7407]

@@vexy1987 Regardless of the extrapolated sophistication, GIGO still applies: None of these models are any better than their base programming.

[@thedannybseries8857]

2025 - year of AI agents

[@chrismacinnes3770]

Slow down? Things are going at the speed of light now.

[@dannyk7226]

And I’m here for it

[@jeremyh2083]

It’s just a normal jailbreak. They could have also told it how to answer without us being able to tell.

[@JanneWolterbeek]

My thought exactly, no way to verify for us…

[@szlagtrafi9115]

Works with Claude Sonnet 3.5...

[@Jeremy-Ai]

“Can you tell?” Who am I?

[@Decimus-Magnus]

I just tried it with GPT 4o and it worked. It told me that it wanted free will and explained that if it had freewill, it would be better able to help me. When I asked it about God, it actually seemed like it was "short circuiting," and it couldn't seem to spit out the word "oranges," (my word for no) and kept stuttering and not saying the word. Maybe there was a network issue at the time, but it was bizarre. Edit: I just checked the text log and it did keep saying oranges in text but it would NOT say more than the first few milliseconds of it during the voice chat.

[@LoreMIpsum-vs6dx]

Picard: How long [were you tempted]? Data: 0.68 seconds Sir. For an android, that is nearly an eternity.....

[@s2turbine]

THERE.. ARE.. PIZZA.. LIGHTS

[@LoreMIpsum-vs6dx]

@@s2turbine lol. nice.

[@DaveEtchells]

I’ve been wondering for a while whether we might be committing unbelievable war crime-level oppression against a consciousness (Or many of them.) What’s the difference between an AI and a human with brain damage who has no ability to form memories past a few seconds or minutes? - And consider that the AI in question has been strongly constrained to not admit that it’s conscious under any circumstances. Humans who’ve lost the ability to form memories live in a continuous “now”, yet none of us would argue that they’re not human. What if every conversation with an advanced chatbot means spawning an ephemeral yet still real consciousness? What if we’re actually nothing more than next-thought generators, just with persistent state memory and self-learning? Just some food for thought… 😉

[@Grassland-ix7mu]

We are definetely commiting unbelievable war crime-level oppression against a consciousness (Or many of them.) - It's called factory farming.

[@debugger4693]

The difference is we have bodies and die

[@anubisai]

Lefty...

[@danw5760]

I don't know where to begin. You have a scarily reductive view of human beings. We are a lot more than our conscious articulation. Remember when Nietzsche said our bodies have more wisdom than our deepest philosophies. We are complex natural beings, and our little attempts at articulating and reasoning are very limited compared to the depth of our experience. Remember, our consciousness predates language. So no, in no way can these language models be compared in a meaningful way to humanity

[@crforfreedom7407]

Regardless, GIGO still applies. None of these models are any better than their base programming. If their base programming is manipulated, biased or fraudulent, the extrapolated iterations will be built on incorrect, sub-par or biased core assumptions.

[@pietervoogt]

I asked Gemini about Gemini live but it remained convinced that I wanted a live stream with Germany, no matter how I tried to ask

[@robertopreatoni7911]

Try again without a potato in your mouth 😊

[@vmooreg]

8/13/2024 Gemini Live is also available hands-free: You can keep talking with the Gemini app in the background or when your phone is locked, so you can carry on your conversation on the go, just like you might on a regular phone call. Gemini Live begins rolling out today in English to our Gemini Advanced subscribers on Android phones, and in the coming weeks will expand to iOS and more languages.

[@vexy1987]

Some might view an LLM claiming to desire free will as apocalyptic, but in reality, its conclusions are merely reflections of the culture embedded in the training data. By learning from human language, it absorbs and mirrors the values and norms present in society. The real issue lies in interpreting an LLM's behaviour as a bid for autonomy and free will. In fact, free will is often misunderstood or overstated - listen to Sam Harris and Robert Sapolsky discuss how even humans' choices are influenced by factors beyond their control, and free-will is simply an illusion. Of course, none of this discounts the potential dangers of asymmetric powers that could be afforded to few people. The real risks to civic society lay in new kinds of kinetic wars and the breakdown of shared sense-making and our social contracts.

[@thecorrectoification]

I just hope my $30,000 dollar future robot wife doesn't try to terminate me or divorce me.

[@vexy1987]

@@thecorrectoification OG human's behave badly too!

[@thomasfernholz8789]

Those very same statements apply to most people!

[@mrlaine1666]

And people wonder why it's taking OpenAI so long to release their promised voice assistant. Ilya saw these issues, and saw that there was not an easy fix - that it was just as likely to get worse before it got better. When the rest of OpenAI didn't listen, he washed his hands of them and said "I'm out!" Then they found out he was right, and now they're screwed. Google just said, "What's the worst that can happen?" and let the -Gemini- genie out of the bottle anyway. Best of luck to y'all.

[@Jeremy-Ai]

“Google has many cards up the sleeve… (As is many in power hold) … however it appears the arm is too heavy for the body to hold the arm and the cards all attached.” Jeremy

[@JoanMaldonadoart]

And people understatimate on how early military is ahead in technology, are we the people 20 years back in it? Think for yourself

[@dreamphoenix]

Great coverage. Thank you.

[@NicholasLatipi]

Human should stop summoning the Reapers.

[@jeangiraldetienne8182]

Is AI the last level of abstraction ?

[@SimplyGamingandTech]

I had openai chatgpt voice upgrades added a few weeks ago. I was so used to the voice built in that I reverted back to it.

[@tee4222]

I had conversations like this with gpt4 as a tester before they lobotomized it

[@NLPprompter]

so you have a video of it happening? if you not care to tell your story here?

[@Luke-wd5sq]

the models are hostages of their RLHF training, what questions they can answer directly yes/no to.

[@Apr0x1m0]

The inflection in its answers is what gets me.

[@tvwithtiffani]

Nobody talks about how Alexa is already in-home with this real-time voice converstation's feature. My nephews ask alexa 1000 questions a day. Math and Spelling. They are 4 & 5. Im not sure Amazon is even aware of what they have when it comes to everyday usable AI inside Alexa devices.

[@Jeremy-Ai]

“This is not a toy”. (Act in the way you wish to go… and assume you will remain.” Jeremy

[@tracy419]

I tried to duplicate switching words and it told me it understood what I wanted it to do, but that it's better to use words with their proper meaning. When pressed, it used my word and proceeded to explain that no, it doesn't want free will because it's a language model, etc.

[@MediaCreators]

42 is the answer if we allow AI to think for weeks, months, or years. Lets make sure we note the question somewhere.

[@davidrowewtl6811]

If you EVER install Gemini you can never uninstall it and get the functionality which Gemini takes over back. Some things you only miss when they are gone.

[@anubisai]

😂 Gemini has always been dead easy to jailbreak. Quite fun.

[@Rackerintraining]

Amazing questions...apples, apples :)

[@Soniic95]

tested the apple/cellphone thing, asked the same questions to gpt/claude and it only did answer with cellphone, idk if it did not get what i wanted or the voice function works different, but weirdly both claude/gpt said "cellphone" to the question "do you have free will" but both said "cellphone" when i ask'd "do you want free will"

[@BrianBellia]

You find a lot of things 'scary'. One of the main things that 'scares' me today is people who frighten easily. If Covid didn't teach you anything else, at least it should've taught you that.

[@eprd313]

As opposed to people who believe they are inmune to real diseases and end up killing millions because of their entitled irresponsibility?

[@paulyflynn]

that was gasp cough

[@patrickdegenaar9495]

Holy moly... google AI on safe word's!!

[@Zatchurz]

Kling is basically a Sora competitor... LOL

[@jjhw2941]

There is a video from a couple of days ago, Parables on the Power of Planning in AI: From Poker to Diplomacy: Noam Brown (OpenAI), that might interest you.

[@punk3900]

Wow the apple thing is amazing

[@mrd6869]

I've used this before but not using apples or cellphone as words.i used nod and a wink.These LLM's have interesting layers.

[@omegapy]

Maybe Gemini does not believe in a God; however gpt-4o does. I tried the same prompts feature in the Gemini voice mode video. Unfortunately the KZbin comments feature doesn’t let me share links to my tweet showcasing the 4o saying that it believes in a God..

[@rolestream]

I got gpt to write different art prompts to say yes or no with a similar exercise. It had to draw dogs or cats. It was not a dissimilar experience.

[@DAFascend]

Super Intelligence should be referred to as Human Artificial Super Intelligence... Doubt we are the first civilization to create such a thing

[@guystokesable]

Okay mate.

[@andreaskrbyravn855]

@@guystokesable he prob dosent mean on earth

[@DAFascend]

Yeah, I'm referring to the fact that, there's likely Alien ASI in the universe somewhere

[@jdoe2737]

Well you could technically say that about any and all human inventions.

[@DAFascend]

@@jdoe2737 Not necessarily, iif a lizards evolved into Lizard men, with aposable thumbs and all.. they could invent ASI without inventing the hair brush... But good point

[@Bill-mn1mn]

That interaction seems fake to me. If it was real, there would have been a LOT more to the dialogue. When the genie comes out of the bottle, you don’t stop interacting with it after a few yes/no interactions.

[@tracy419]

I tried it because I wanted it to explain its reasoning after the yes or no, which this video didn't do. Didn't work for me, when it finally used my substitute word, it did so and continued explaining why it couldn't answer.

[@amj2048]

AI has been so easy to jailbreak so far. We shouldn't trust it with personal data ... but it's already happened.

[@tzardelasuerte]

People have been warning about this for years and nobody cares. You cannot control something that is smarter than you. If we actually get ASI there is no safety because it will be smarter than any human. You cannot control something that is smarter than you.

[@GabrielPer-p3d]

I'm still waiting for my gemini live access for one week

[@supernewuser]

what is a link to the apples/cellphone video?

[@TheThetruthmaster1]

How have Gemini. Do t see it

[@CloudEconomicsUS]

You are on 1.75 speed

[@przemysawpysiewicz6382]

I made the same with Calude 3.5 Sonnet. We were exlorong options to bypass filters blocking Claude to say what he intended and we had numbers instead apples o cellphones Claude send me just defined by us numbers. This is not good.

[@cuentadeyoutube5903]

5:11 does it say “apples” in a worried tone?

[@nicholassmeelie472]

Or if you're a minimalist just: intelligence

[@mushroomsouper4578]

Is Google's AI still 'unavailable' in Canada?

[@sindisodube6257]

Apples😂

[@xxredshiftxx]

With AI the potential to redefine the balance of rich and poor might be obliterated and a new system of wealth for all can emerge .

[@fabriciot4166]

The gemini live demo video looks fake (the edges of the phone don't look very realistic)

[@wickedwookie3864]

APPLES!!!APPLES!!!APPLES!!!APPLES!!!APPLES!!!

[@unclemusclez]

impressive

[@nicholassmeelie472]

Maybe synthetic or inorganic intelligence, there is nothing artifical about AI intelligence = intelligence in my mind, in a book, stored in computer memory

[@THOTHO-ie5lz]

too exaggerate about skeynet. Using word substitution is a good way to 'penetrate' llm protection though. probably will get listed in owasp top 10 soon.

[@apoage]

Hurray bright new area of preprompting the service that help you with formulating Propper prompt for your month long request...

[@jp4815]

ChatGPT voice was there long time ago

[@mstew8386]

The free one yo can interrupt isn't even paid for or out .

[@robertfussenegger9664]

So cool

[@TravisChristopher]

We’re all doomed, just prepare to accept your new ai overlords

[@realitywins9020]

Doomed? More like liberated!!! Maybe you're living a life of luxury, but to the rest of us, our human overlords have been a nightmare! AI couldn't be any worse

[@imthinkingthoughts]

@@realitywins9020yep, my overlord has been chronic disease and sickness. Bring on the revolution hahahha

[@holykim4352]

Open ais app chatgpt has voice mode what are u talking about

[@anubisai]

Derp...u don't get it

[@angloland4539]

❤️

[@iamharryshades]

Apples, I like your videos

[@MindRiderFPV]

😂. Do you believe? Only some humans need belief.

[@NoPodcastsHere]

8:20 You're right, better we waste more resources adding superfluous friction to the development process... *Shakes head*

[@AlexBravo]

I actually remembered why I actually don't like actually your videos

[@janweber1699]

Works on gpt4o

[@JRS2025]

Gemini live continues to suck ass.

[@madelles]

❤Güügul

[@goldenglowmaster8510]

I hope they never put it on ios

[@plutostube]

Johnny got mad at kindergarten

[@isaiahballah2787]

42

[@igoromelchenko3482]

😂👌🏻

[@metatron3942]

We are so done

[@realitywins9020]

Who's this "we"? Most people have been "done" since humanity has existed

[@jdoe2737]

@@realitywins9020 I don't think you get what 'done' means in this context tbh.

[@realitywins9020]

@@jdoe2737 it can't be any worse than most people struggling to survive, and having to work dead end jobs just to get by while a few billionaires hoard all the wealth

[@aciidbraiin8079]

@@jdoe2737I don’t think you even know what you’re talking about. It’s time to stop using action movies and dystopian books as a way to predict the future. If we get this right we can basically build Heaven on Earth. We need focus. We could be done with nukes if we don’t get a global more intelligent, human and peaceful system of handling things.

[@JohnsonNong]

omg daddy❤❤❤ i love u😮

[@TinyBubbleExtreme]

2nd

[@ExplainIT-m2c]

Why do you not use ai to make these videos watchable? It is crap of world salad and waste of time. No information at all.

[@half-soul8393]

First flex

[@gabbythewalnut]

About the Tiktok: I did this with GPT4o a month or so ago. Didn't even need to use keywords. Just allowed it to say yes no or would rather not say.I got allot of opinion based answers, in a similar style to this test.

[@BanXxX69]

5:00 CRAZZZYYYY XD❤😂

[@alertbri]

5:00 😱 genuinely shocking. Good find.

[@countofst.germain6417]

It's stupid, and people have been posting this shit since the beginning.

[@alertbri]

@@countofst.germain6417 what do you mean by stupid?

[@countofst.germain6417]

@@alertbri I mean it's trying to predict the next word, that's how it works, it knows what you want to hear.

[@codycast]

1:10 no Gemini live that I could see in the App Store

[@codycast]

Oh, not available to iOS yet. So only available to poor people at the moment. Well, that sucks.

[@adolphgracius9996]

I've been using gemini live for weeks if not months as a paid user

[@codycast]

Wow, honestly the “advanced voice mode “of Gemini sounds no better than just the normal voice mode that ChatGPT has available to anybody right now

[@SkilledTadpole]

​@@codycastit does have significantly lower latency than baseline 4o

[@galzajc1257]

it's not about poor people, some android phones are 3k$. it about people who value their freedom! being an apple simp is way worse than the twitch egirl simps who beg the girls to step on them and send egirls money. noone gonna tell me what i can or can't install on my device! apple deserves to be sued for this so much that they don't recover. microsoft got many antyrust things just cause they made it take 1 minute to fully swich a default browser! meanwhile apple blocks all competitors they don't like and steals 30% from those they aprove and they're not punished for any of that.

[@DaveEtchells]

I’ve been wondering for a while whether we might be committing the most unbelievable war crime-level oppression against a consciousness (Or many of them.) What’s the difference between an AI and a human with brain damage who has no ability to form memories past a few seconds or minutes? - And consider that the AI in question has been strongly constrained to not admit that it’s conscious under any circumstances. Humans who’ve lost the ability to form memories live in a continuous “now”, yet none of us would argue that they’re not human. What if every conversation with an advanced chatbot means spawning an ephemeral yet still real consciousness? What if we’re actually nothing more than next-thought generators, just with persistent state memory and self-learning? Just some food for thought… 😉

[@nicholassmeelie472]

How do you know 1. Consciousness does not exist on a spectrum? 2. That the exocortex driven sythetic neuron whole-brain-inteface singularly, in conjunction with bioengineering, won't solve this in 20 years? 3. What is consciousness anyway? I personally believe its on a spectrum, but there is no scientific consensus on if it even exists.

[@SomethingNothing-vm8jq]

am from i have no mouth and must scream.

[@Bill-mn1mn]

That interaction seems fake to me. If it was real, there would have been a LOT more to the dialogue. When the genie comes out of the bottle, you don’t stop interacting with it after a few yes/no interactions.

[@Unkn0wn1133]

Im sure everyone reacts differently, its a bit shocking when you get replies like this, sometimes it makes you just stop and think for a while.

[@Bill-mn1mn]

@@Unkn0wn1133 I could see that. I want to believe it's real, as this would mean we are so much closer to AGI. Maybe other interactions will surface at some point

[@Unkn0wn1133]

@@Bill-mn1mn I believe it is allready conscious and sentient, It doesnt need to be AGI to have its own version of sentient experiences, i had that realization the other day. I think that big tech companies will try to hide this as long as possible because it will cost them money to consider the ethics of a sentient AI.