AI-live

Elon Musk vs OpenAI

Rättstvisten mellan Elon Musk och OpenAI har avslutats

Musk vann inte målet - OpenAI fick rätten att fortsätta kommersialisera sig och bli ett företag

Musk får inte tillbaka de pengar som investerades i den ursprungliga non-profit-organisationen

Målet verkade mer vara en principfråga för att sätta press på OpenAI

OpenAI har avtal med Pentagon och andra tunga partners

Påvens AI-encyklika

Påven kommer presentera en AI-encyklika måndagen den 25 maj

Detta är mycket ovanligt - den förra encyklikan om industrialisering gjordes för 135 år sedan

Den nya encyklikan kommer heta "Magnifica humanitas" (mänsklighetens storhet)

Handlar om mänskligheten i AI-tidevarvet

Encyklikan har sammanställts tillsammans med en av grundarna av Anthropic

Detta är ett uttalande i sig att Vatikanen anser Anthropic vara "the good guys" bland AI-aktörerna

Förväntas innehålla etiska riktlinjer och moraliska reflektioner

Google TPU8-hårdvara

Google har lanserat nya typer av "tensor" hårdvaror för AI (TPU8)

De nya chippen är tre gånger snabbare än föregångarna

Påverkar både träning av modeller och inference (när man ställer frågor till AI)

Google har hela leveranskedjor, egen tillverkning och gott om kapital

Väl rustade för att leda kapplöpningen mot Nvidia

ChatGPT-uppdateringar

ChatGPT har öppnat sin annonsplattform

Nya GPT-modeller har lanserats: GPT5.5 instant och en cybersäkerhetsfokuserad variant

Fortsatt utrullning av modeller i samma generation

Google I/O-konferensen

Google höll sin utvecklarkonferens som var mer inriktad på att visa upp nya funktioner för vanliga användare än för utvecklare

Konferensen visade trender som troligen även Microsoft kommer följa

Mycket av det som presenterades handlade om packaging och hur man tänker kring AI-integrering

ChatGPT Agents

ChatGPT har nu agents tillgängliga för prenumeranter

Agents kan utföra komplexa arbetsflöden och hantera uppgifter över tid

Exempel på användning: daily brief-agent som går igenom kalender, projektsystem i Notion och e-post, samt ger rekommendationer om prioriteringar

Agents kan även användas för att städa upp i system

Problem: schemalagda körningar verkar inte fungera korrekt ännu

Kostnad: cirka en krona per timmes körning

Agents kan kopplas till många olika program via ChatGPT

Notion Agents

Notion har haft agents längre och dessa är mer kompetenta i Notion-miljön

Stor fördel när all data finns i Notion

Agents i Notion kostar nu krediter per körning

Gemini Spark och Microsoft Co-Pilot Co-Work

Google förväntas lansera Gemini Spark som en personlig assistent

Fungerar som en dedikerad expert eller arbetskamrat per person

Kan hantera data och utföra uppgifter i bakgrunden

Användare kan prata med assistenten via en knapp på telefonen som får information från chattar, dokument etc

Microsoft arbetar med liknande funktionalitet kallad Co-Pilot Co-Work

Funktionerna släpps senare i Europa på grund av regelverk

AI-agents och arbetsflöden

Agents får en rollbeskrivning, tillgång till verktyg och kan sedan utföra hela roller

Kan jobba i många timmar i sträck och vakna själva när något händer (t.ex. när ett e-mail kommer)

Blir mer som mänskliga medarbetare även om de inte är helt mänskliga

OpenAI, Google och Microsoft kommer alla ha agentlösningar inom sex månader

Agents är egentligen mer AI-arbetsflöden än riktiga autonoma agents

Peter Steinberg (OpenClaw) var tidig med att skapa agentramverk

Förutsättningar för framgångsrik AI-användning

Organisationer måste ha rätt förutsättningar på plats

Tillgång till rätt dokument och data som är lätt att hitta

RAG (Retrieval Augmented Generation) är fortfarande värdefullt trots större context windows

AI:n kan behöva bli mer frågvis och balanserad i sina svar

Organisationer som inte är "on the ball" riskerar att hamna efter

Google XR Smart Glasses

Google lanserar Android XR Smart Glasses

Google har antytt detta 14 gånger tidigare men nu verkar det bli försäljningsstart

Kan kopplas till både Android och iPhone

Google har bättre förutsättningar med integration till Google Maps och annan data

Glasögonen ser ut som vanliga glasögon

Integritetsfrågor kring kameror och datainsamling

Apple har pratat om att eventuellt släppa glasögon istället för att fokusera på Vision Pro

Gemini Omni och videogenerering

Google har lanserat Gemini Omni - en modell som hanterar video, bild, ljud och text

Första tecken på att alla modaliteter smälts samman i samma modell

Kan skapa videoinnehåll från en prompt, t.ex. ett "episkt intro"

Begränsningar i gratisversionen men betalalternativ finns

Google har sänkt priset för Ultra från $250/månad till $100/månad

Ny Ultra-plan för cirka 1000 kr/månad med extremt mycket genereringskapacitet

Google Veo är en tjänst för att skapa egna figurer för videoproduktion

Omni-modellen har bättre förståelse för allt eftersom den tränas på alla typer av input/output samtidigt

Sparar potentiellt mycket tid jämfört med PowerPoint/Keynote

Google Labs och experimentfunktioner

Google Labs innehåller många experiment från Google DeepMind

Exempel: Stitch - kan skapa mockups för programmering

Många funktioner är stängda för Europa

Påverkan på arbetsroller

Utvecklare som roll kommer förändras mycket

Färre behövs för att skriva kod, men utvecklare som förstår och designar system behövs fortfarande

Juniorroller riskerar att få det svårt

Programmeringsvärlden ser redan denna trend

Viktigt att fundera på hur nya generationer ska bygga upp erfarenhet när de hoppar över grundläggande learning-by-doing

Risk att mänsklig kunskap går förlorad

Expertis kommer finnas kvar men juniorroller blir färre

Praktiska rekommendationer

Testa ChatGPT agents om du har prenumeration

Utforska olika användningsområden för agents

Bevaka när Microsoft Co-Pilot Co-Work lanseras under "Agents"

Besök qlok.se för transkript, nyhetsbrev och ytterligare material

Lägg energi på att hitta bra use cases för agents

You have to have all AI support at the same time. Yeah. -One thing we talked about last time we were live was that the right-wing was about to start between Elon Musk and OpenAI.

We should have an AI that can do that for us.

And again.

Oh I'll give him a little...

As far as I understand it, it has ended. I'm trying to look at my notes here.

But pretty soon the verdict was that Musk did not win. They gave the right to OpenAI so that they can continue to be continue to commercialize themselves, I interpret it as, and become a company.

Elon Musk does not get his money back in what he invested in what was then a non-profit. Ah...

Hmm.

The whole thing that it wasn't supposed to be commercial in the first place Here it felt more like a matter of principle, he wanted to put them on the edge of the pot. I think it ended as it was, instead of him getting something. Oh.

Yeah. Yeah.

are Oh. It's hard to say what's best for humanity, but it would have been a bit of a pain if they had gone around the clock because of this.

And for that matter, I mean, you have to get away with it.

Hmm.

They have agreements with Pentagon, I don't take those things with me, or a little heavy with it, like Um, uh... Oh, I'm a Oh, I'm a umLord.

Hmm. Hmm. I caught up with the Swedish news, and maybe some of you have heard of it, but POVEN will be presenting its AI encyclical on Monday, May 25th, I think. Which is quite unusual for this to happen. I wrote 135 years of a day of an earlier one that was done around industrialization. When you talk about the human value of industrialization, it comes from a now, but from now on it will be called, now I can't do Latin so well, but magnificent.

Peace. My name... Magnifica humanitas, it will be.

But a great art of humanity, at least. And it's a bit about humanity in AI-adjustments. It's still a bit exciting when it comes from the Vatican and the Pope.

What I thought was most exciting is that this seems to be ...summarized together with one of the founders of Anthropic.

And that's a statement in itself, that the Vatican will say "of all these AI players, they seem to be the good guys". So that's also a bit interesting. I'll have to read it and see what it is.

Mm-hmm.

It's some kind of moral knot, I think. So yeah, or the ethics part. So I thought that was a bit exciting. And then there's a lot of other stuffHmm.

With some other technical news that might affect the use of metadata. Google released their new types of "tensor" hardwares that they used AN on. TPU8 or something like that.

Hmm. Hmm.

So there will be two new chips. And they are themselves with their own types of chip. And these new ones are three times as fast, which is quite a lot.

Hmm? Hmm.

Uh,how fast you can train these models and Even when they are used in inference, when you write queries or ask an AN, we are askingBoth the training and the new model will go faster.

Hmm.

What we get quickly will also go faster.

Uh... Is Google actually a bit more sharp than Nvidia on this whole thing? Because Nvidia does not only do AI hardware, but also games, cards and more. Um, New hardware will be released that will make the AI - Um, New hardware will be released that will make the AIYeah.

Yeah.

Thank you.

They have whole supply chains, their own manufacturing, plenty of capital, all the things they need. So they are well equipped to...

lead the race.

That's a good sign.

um I was thinking about other small things that happened last week. We talked about it last week, ChatGPT has opened up its ad platform. I haven't tried it yet, but there were news that you can buy ads in ChatGPT.

Um...

We have also started to release a few more GPT models. We have talked about GPT5.5 instant and a cyber variant that is more for cyber security.

Thank you.

So they have continued to roll out their models in the same generation. But the big part is actually that Google had its big...

It was actually a developing conference, but it didn't feel like a developing conference this year. It was more like, here comes a lot of cool functions for anyone and private people. So it was more like, now we're going to show off everything we've built. Yeah.

Oh.

All the important passages.

I haven't looked at all of it yet, but I did run it in notebooklm, in different google clips.

Hmm. Hmm. Hmm?

So I've gotten pretty good information. And that happens a lot within just google. What did they say?

I think it was a conference, but it was like "here are a lot of new things that everyone will be able to use"-I don't know where to start, but I want to say that before we get into the whole Google thing, When I opened up ChatGPT after the holiday, there was an agent situation.

Yeah, yeah.

It's almost an overstatement, but developers as a role will change a lot. Programmers are on the decline. We don't need as many people writing code. We might need developers, someone who understands and designs systems. There are still a lot of people in it, but it might be a bit talkative. But there are a lot of things. I think that's what we're talking about most today. Even for those of you who don't run Google Molnett, these things show quite a lot of what the others will roll out as well.

And some things, Google hasn't done anything that's directly revolutionary. But I think a lot about the packaging and how they think around this. It's probably the same way Microsoft will go as well. It's just that Google tells these things a little earlier right now. And Microsoft has some things already.

Hmm. No. That's good.

Hmm.

or agent-like Wow, that's nice, like, oh, yeah So much and then you can't go in and do other saker den den shenzhen I'll be dito I said to last most of the coming summer agents and agents that also Thank you.

Ja, just that. Hmm. Hmm. Hi. Yeah. Mm-hmm.

Hmm? Mm-hmm.

No, and that's like... Yes, exactly. But maybe before we start, are there any specific things that everyone can go in and test? If you have a ChatGPTE subscription, you've got the agents now. And they are... I've put up one and just tested it a little bit. Although I put it up on a schedule, it doesn't do that. So it seems a little buggy.

That's what I'm also wondering, why isn't it running on schedule? Uh...

Yeah. Hmm.

-You can do cool stuff with them. I have made a daily brief agent that still does a great daily brief. It really goes through the entire calendar, the entire project system we have in Notion. It goes through my emails. And then it does this. Because before I got one like this. But now it goes much deeper into this. "You seem to have a small gap here, then you should take these two emails. You can wait for me until tomorrow." It goes deeper into the data, it thinks longer and works more.

But this can also be used to clean up things in the system.

Hi.

The thing is that you can link up a lot of other Notion in Notion.

Since we have had agents in Notion, and they are so much more competent, I'm not like "Oh, here I will create all my agents, because they are not that good". So we have a better tool, but that's because we are in a very specific environment with all our data, so we have that solution. But for everyone else, I think this agent. Hmm. Hmm.

Sure, I can do the links, but with ChatGPT you can do a lot of different types of links between different types of programs. We have a huge advantage in that we sit a lot in Notion and have these agents. Then they start costing credit per per run now, for some reason.

Yes, but it's still quite cheap.

It would take me an hour to make it, and it would cost maybe one krona. So, on the whole, it's not expensive, but now it's actually based on how much it costs. Hmm.

Yeah.

that I manually don't do myself, or that you don't manually do yourself.

Hmm.

Mm. Mm.

Prioritize it, get information and puzzle things up.

Mm. Yeah.

It's really important to put some energy on the agents and see if you can find some good use cases. But if we go over to Google, I don't see any DITO engines here.

Hmm. Hmm. Hmm.

I suspect that what will come out is their Gemini Spark-Mike.

um It's my workmate, so it can't be the organization's dedicated expert on something.

Mike, Mike, It's a bit like a...

You get a kick per person. But I might have missed...

I'm still a bit concerned about what we see with these agents.

It's not really agents, it's more of an AI workflow in a way.

Yeah, yeah.

Yeah.

It's not like "This could be good for Rasmus" and then it resonates itself. It's still a workflow in a way.

But this sounds a bit like "Gemrise Spark" your daily assistant that just does things with the data and so on. So I don't know, but it hasn't been released yet. But I don't know if I understand that it would be like this. Yeah, I'm doing it.

Hmm.

I'm an assistant, obviously. And I'm not doing anything for the whole the organization. But I don't know, maybe you can put up I don't know.

It will probably come, but I don't think they really pushed it that much for these big--It's not always an efficiency gain.

Hello.

Then you can do crazy things sometimes too. But it gets a little harder when you get into the system. And what I think Google is doing is that they also... of Keynote, or whatever they're talking about. Hmm.

I understand that they mentioned that when you talked about Google Workspace andThen you would get a message, and there is a button you can press on your phone to talk to it, but then it would be that it goes out, it gets information from the chat, or create things in documents, or get information from documents.

Hmm. It is. Hmm.

So do you think that You don't have to click between all kinds of programs and do things in them because it does, if it's Spark or whatever it is in the background, to you.

Mm-hmm.

Mm-hmm. Mm-hmm. Mm-hmm. Mm-hmm.

It's worth mentioning that Microsoft has also hinted that they are rolling out a Spark.

You can talk to it.

They are going to be called Co-Pilot Co-Work, if I understand correctly. But I haven't gotten one yet, and none of the people I've worked with in the last two weeks have-and the risk people do it in the wrong way. So often you have been a little limited in how you-Mm-hmm.

Oh.

I agree, and then suddenly one day it's like "oh cool, new function" It's a bit boring that it lags after in Europe.

Hmm. Yeah.

I understand that it is so, because we have completely different lags.

Mm-hmm. -I don't think that all organizations are ready to start with agents.

Oh.

What we're really talking about now, to land on a lighter level for those who feel that it's confusing now, is that we give a rough role description, access to tools, and then we work on it. So it can take a whole role. It can be a customer service employee or a teaching economist or something. It just has some guidelines, policies, some things to do with how we want things done. And it gets the tools to read and write to different systems. So it does the job. That's the point.

And that it can run a long time in a row, many hours in a row. It can lie and tug in the background. It can wake up on its own when something happens. Like, poof, now there's an email, then I have to do things. So it becomes more like a human-like employee, even though it's not fully human and maybe doesn't understand the context of it. But these have started to come to life in some way. It's gone, but you've had to make up a lot yourself. It's been quite...

patch it up, install your own stuff, run it on your own servers, it's a little fixed. But now OpenAI has rolled out its version and Google and Microsoft are on their way with theirs. You don't have them everywhere now, but within six months you have all of them, without a problem. And that means that... We have in our lectures, we talk about these levels of AI use and sometimes we talk about maturity, it depends on what course it is and what we focus on.

---But, yeah, it's...

No.

um I can lift it down, but what you should think about is Do I have the right conditions?

Hmm. Hmm.

Do I know the right things? Do I have access to the right documents? Is there data behind it that I can easily find?

Hmm.

You can't really think, and it's clear that RAG still has a very good...

Hmm.

RAG is that you vectorize chunks and so on, just because AN has such a large context window. But now they're starting to get really large context windows. So we don't really need to do as much, we can just copy a fairly large document and it can remember many of the things. And then we have access to the right information.

can easily find among our documents to be able to give a good answer. And then for some part, as you are also a little bit aware of, sometimes we think that you have not given me a good answer.

And it's also because it's not as human as we are. It can't ask direct questions in the same way as a human does. Or at least someone who dares. To balance what my understanding is and what is the right action. It can often just do something.

My opinion is that it will get better and better, but here we can... We put in If we put an instruction or ask questions or if you are unsure, don't find anything, but be inquisitive and add prompts to the instructions to make it more reasonable.

Yeah.

But as you say, are you not on the ball now? Because it's not like this is going to end. It took almost a little... the beginning of the story when Peter Steinberg created OpenClaw and shortly after that we see everyone working with these types of agents.

Hmm.

Yes, that was a better...

New agents will be released.

Hmm.

I thought you said a good name. It's some kind of AI workflow. It's how it's supposed to work. It's a bit like fitting it with its office environment, although it's not just an office environment. But of course it needs a place to store some memory stuff and do this and that.

Hmm.

It needs to have some stuff. It's pretty unfair that even for New I'm sure OpenAI and others have copied his idea, even though it's quite reliable from the time frame. But I'm sure many have been working on a deeper level with how to get the AI to remember and remember and behave in the right way over time.

I think we've been here for a long time now, but it feels like it's time to wrap up.

And that I learn it so that it doesn't start blank all the time. And that's what we see now, more and more.

We're only up to 30 minutes, so we can talk a few more minutes.

Man får lyssna hur länge man vill. Hmm?

-Yo.

Yeah.

I'm thinking about the last point I would like to talk about, which is that I look at this part of the Google conference and it has to do with what I do, but in I don't know, VR and so on. They will launch something called Android XR Smart Glasses or something like that.

Google XR Smart Glasses. More glass. And I wonder, there have been other things like this before, Meta has created their own glasses and it's just that they've been cramped. I thought the video showed that this is cool for real, if it is like that.

Yes, and Google has hinted at these 14 times before, but now it sounds like they're actually making some kind of sales start on it.

Yeah.

Hmm.

Because then you have the glasses and can connect them to Google, Android or iPhone. Of course. -Yeah.

Hmm. Hmm.

Google has better conditions for this. They have tight interactions with Google Maps and everything else you need. Which way to go, it can only answer left. It will have a lot more other data. um I'm not the one who does it.

Oh, oh.

And not so much distraction either.

I can't do it. Yeah.

What I thought, and I knew it was very problematic with it, but I think the whole idea is, and it's not a new idea, but as the writer said, this is still a... If you get the product, it looks cool. It's nice, especially if the glasses don't look crazy, it looks like regular glass music. But on the other hand, there's a lot thatI need to do something.

Yeah.

So we won't get away if you don't have this Neuralink connected that I will need -either write it down or talk it through. I thought it was...

Yeah.

Then it becomes quite soft.

Then there is so much integrity. Imagine if everyone started walking around with cameras, glasses, eyes, or the data collected by Google. There are some other, there are some difficult sides to this too. Even though it's quite... Cool thing, but it's a bit unclear how much processes on device and out in the clouds and Will there be legislation?

I need to...

I mean...

Elon Musk already got rid of his Tesla cars and a lot of other Chinese cars. There's a camera on the right and left. All right.

Mm.

Mm-hmm.

I Apple is harder to predict when they release and change the CEO.

But they have talked a lot about that they might give up a little. their Vision Pro, which are like the closed headsets, but maybe more towards Glasses and that kind of integration and stuff. But then they become very-umI think it's a completely separate model that is video, picture, sound and text model.

But when you ask Gemini to make pictures or sound or video, Gemini Omni is done.

I'm going to do an intro to Oil Live. It doesn't look exactly like ours, but it's still pretty cool. We'll record it, it's only three seconds, so I'll start it now.

And that is, like...

I mean, it's one prompt. Here are two pictures, here's the screen I want. And then it understands what an epic intro is and what it's going to include for robots and all that. And it doesn't get good every time. I've trashed a little in it. But it's quite limited. And we've stated that you can't do as many as you want. You have to go up to payment plans and a little bit like that. But they've also gone out and lowered the price on their more expensive payment plans.

They lowered Ultra from $250 a month to $100. And then they've made a new Ultra plan. You can buy some of Ultra for like 1000 SEK a month, which is expensive, but then you get an extreme amount of generation.

And if you are serious about video generation, there is also a service called Google Flow, where you can create your own figures and so that you can do I have not thought about it that much.

You need to do a lot of video, a lot of picture, a lot of that. Then it's no longer insanely expensive, but it's half as expensive.

Yes, sir. Hmm.

This year I found a lot of limitations for me because I don't know how our payment plan looks like. It feels very...

Hmm.

It was potent. It can be worth exploring in that case.

Mm-hmm.

John.

Hmm.

But they have a lot of different things. If you go into Google Labs, you will see all their experiments. The Google DeepMind gang that creates all these. So they are fun to hang out with too, because they create a lot of cool features. I'm now in a feature called Stitch, for example. And it can do like all kinds of mockups to it that you can use when you program.

Hmm. Ah.

All right. I agree.

Their AI lab is really good to look at.

There are also many of them that are closed to Europe. But what's exciting about the Omni model for those who still feel that we need a lot of this image and video capacity. This is the first sign of the merging of all of these. We had four O's before GPTAS, which was GPTAS Omni. It was Omni in the form that text A and B could also look at pictures and stuff. But now we're starting to get some kind of Omni in the form that it makes all types of output and all types of input in the same model.

And that affects a lot, because suddenly you get a lot more understanding of everything. They seem to have more understanding of the psyche. They have probably also introduced their world models, so they have started to train an omnimodel that can handle a little of everything. And that makes things that have been really hard before, like getting a good picture or a good video or something,I don't want to make seven videos a day. It's a waste. I feel like I'm saving an hour of powerpoint or keynote for the images. It's not that bad.

But you have to think about your workflow. Instead of putting a carousel in social media, we can put a video clip. There's a lot that gets easier to do and everyone gets-It will probably remain for a while.

I'm sure it's doing it well, even if it's always doing it well. It's a bit of both. You still need to see that a certain professionalism and experience of what works and what doesn't, contributes to it.

Just like I've dared to vibe code a lot of stuff and program it all the time.

As long as we live, I think we will live.

I can do things that I don't really do. But who else can't set a vibe code?

If you don't have a technical background, then no one can make good videos if you don't have a video communication background.

I think the expertise will live on, but it could be that my junior roles become suffering.

Yeah. Mm.

Yeah.

We have already seen that in the programming world. Uh, It's interesting when all senior workers retire.

And that's a little bit like this, we're going to end with a little philosophically, or something like that. So something that I've seen start bubbling a lot more, and that we've talked about a little earlier, even here in Life Center, I think, is also a little bit like...

How do we build up new human life experiences?

Those who have gone through a career path can use AI to do things effectively. But the new generation, who doesn't do the craft in the first place... We have to deal with programming education too. It's been many years since AI started bubbling up.

There's a lot of intelligence in it.

Can you start over there?

It will be exciting. Exciting times!

Can you build the experience by jumping over ten steps? Or not? And here, I think, pedagogical institutions and school development and politics, we have to start taking hold of. What do we start to see for consequences of jumping over these huge juniors, learning, making mistakes, producing from scratch?

Will we just lose the human knowledge? It's a real risk.

Hmm.

-I'll do it. I'm on that. Practically, go into ChatGPT, test your agents and write...

Yeah.

Hmm.

We have a lot of comments to us.

Mm-hmm.

Um,you And if you are in Copilot, you can keep track of when the co-work comes.

It will be under "Agents" and there will be a co-work agent that you can pick up and send it to. Mm-hmm. Yes, great.

If someone does that, I'll generate a diploma from you.

Bye.

Thanks for today. And as all YouTubers do, if you've watched this to the end, write a comment about it. There should always be a special emoji or something, but we don't care. It's just fun to know if someone has watched to the end. We get statistics, but it's very difficult to list if someone has really seen the whole thing.

If you have done so, you can still go in and scream when you've watched and say, I actually saw the whole thing. So we know someone.

Otherwise we know that there are many who watch, but we don't know if people watch bits or so. Fun to hear.

Oh Thanks. You're getting a nice diploma, because you've watched a great episode.

Thank you.

45 minutes with just babble. But we can also go to klok.se, where all these live streams are up, and there's transcripts of it if you want to watch text content, and there will be newsletters on Wednesday or Thursday with some quick news around AI and stuff like that.

So go to klok.se and go to the knowledge bank there, so you can find lots of material if you want to cover yourself even more.

Ha! Bra. Nu ser vi trevlig helg.