Microsoft`s New INSANE AI AGENT Magentic-1 - TESTED

[@MichealPeggins]

Would you consider doing an installation video of Magentic one? I tried on my own but i guess I'm not proficient enough with coding. I couldn't figure it out.

[@themax2go]

about the summarization of your yt vid for 1min... maybe it added the time since initial "play" - which was the first ad? meaning, it started counting the seconds since the first ad started?

[@mrpocock]

So it seemed that it did not follow your instruction to test and run the snake game.

[@themax2go]

thing with prompting (right now) is, unless you implement proper prompt engineering (and "being very specific" in your prompt is one part of that) current AI doesn't do that great - think of putting in the same request to a "not very enthusiastic" (aka not experienced / professional) freelancer (or you're not offering enough pay), it'll give you the same results. think in line of: the amount of effort i put into prompt eng == monies offered for a freelance gig

[@khariclemens]

That’s actually ok. Imagine deploying these for small or home based businesses that typically pay an intern. Now they can pay you.

[@themax2go]

@@khariclemens hmm i don't get your point... explain?

[@khariclemens]

@@themax2go it’s fairly simple. This tool will only improve over time. As of right now, we are able to carefully provide a very specific prompt that can allow this agent to complete simple jobs that would normally be ‘time intensive’ but not ‘difficult’. We pay interns for things like that all the time, and the jobs are ongoing. As a developer, I can analyze the task, write a good prompt ONCE, and deploy this thing, effectively getting rid of the need for the intern AND their salary. So if the client was paying that intern say $1000 a month, I can charge half of that and get paid monthly, or I can charge a higher ‘one time’ fee and the client never has to pay an intern again. If you ask me, that’s pretty powerful. I’m working on one as we speak.

[@anieve01]

dude, your videos are GOLD... much appreciated! looking forward to seeing more

[@goethe528]

Can it find emails in Outlook?

[@tomaszzielinski4521]

Wow, I'm really impressed that AI got through KZbin adds without going crazy o.O!

[@alphamindset9634]

Chapters?

[@miad8707]

This is interesting. I'd like to try it with OLLAMA. Do you have any information on how to do that?

[@MrErick1160]

So this is actually better than Claude’s computer use

[@khariclemens]

Wait a minute. I know this isn’t on topic, but I’m willing to bet the DOGE acronym was Elon’s idea.

[@Craig_Doll]

Genius, how'd you figure that out

[@khariclemens]

@ a little bird twittered in my ear.

[@YT_Jx]

@@khariclemens r u sure you didn’t hear it from your X?

[@khariclemens]

@ lol. Nice.

[@IsZomg]

Someone else on X suggested it and Elon said 'thats pefect!'

O youtube traduziu para português o audio do seu vídeo. Mas ficou muito ruim!

[@aivy-aigeneratedmusic6370]

thanks for this video it looks very interesting and promising. i wonder how many tokens does this process use? and which ai model does it use? sorry if you answered this in the video, i jumped directly to the tests 😂

[@JohnBoyle-xp6qu]

github says gpt4 only for the moment.

[@llucis-v]

@@aivy-aigeneratedmusic6370 no, this info is lacking, not sure why he decided it's not of interest 🤔

[@SebKrogh]

Any low-code agent / automation platforms that can use these?

[@StefanYT2810]

Der Ton auf Deutsch ist schrecklich.🙈

[@themax2go]

BEWARE HOW YOU ASK, IT DEFINES WHERE THE INFO IS BEING PULLED FROM => "claims to perform...": benchmarks mean squad (anymore). qwen2.5 (32b) "sucks" (right now) @ instruct following and coding - it seems chinese ai companies (qwen is run by the alibaba group) "tune" their models to do well on those test cases / leaderboards, but they otherwise don't do well (very badly actually, inconsistently) - ps. insight abt "chinese ai companies" comes from someone who seems to run a chinese ai company.

[@llucis-v]

total cost for these tests?

[@jarek_rudnik]

If I understood this correctly - electricity and internet costs (probably also huge environmental impact). This one is pretrained and you install it on your PC. It does not hit tokens for external services. It is like Ollama models which you have on your PC/Server and can interact with them even without internet connection. I actually wonder what are minimum requirements to run this agents on your device to get nice performance.

[@llucis-v]

@@jarek_rudnik nah, it requires OpenAI API key, uses gpt4o for the agents, multimodal stuff etc. it's annoying that the video glosses over the API requirements and cost here

[@jarek_rudnik]

@@llucis-v yes you are right: Currently, Magentic-One only supports OpenAI's GPT-4o as the underlying LLM. Cost is not mentioned but I am sure it is way to expensive :) for what it does. I hope they plan to expand this for other models but let's see.

[@themax2go]

? local models (ollama / lmstudio / ...) ?

[@xmagcx1]

very bad traduction, pls change 😂

[@AdvantestInc]

This exploration of AI agents really highlights the exciting new directions for tech. Watching Magentic-1 tackle web searches and video summaries opens up so many possibilities for time-saving in information gathering. Looking forward to seeing how multi-agent AI could evolve beyond research tasks!

[@avataros111]

a finger from me too

[@extremelylucky999]

Not to be that guy, but when I google about Elon musk new role, I can get the information I need like 1 billion times faster than this. I don’t understand why this is useful.

[@khariclemens]

@@extremelylucky999 you’re being that guy.

[@alphamindset9634]

I was thinking the same thing lol beta version w data collection

[@jarek_rudnik]

Look at the trend. When first vehicles were created people were saying: "Not to be that guy, but this is quite slow. I can walk faster. I don’t understand why this is useful". It will get faster. It will get smarter. It will get cheaper(?). It will (probably) ruin the environment. It will definitely make a change (I am not sure for better or worse).

[@extremelylucky999]

@jarek_rudnik I have to disagree with your analogy, though. Firstly, the first motor vehicle traveled at 3 to 4 times the average walking speed of a person. Next, the demand for vehicles only increased commensurate with the rate of industrialization. Demand from ordinary people that is. It’s tough to see ordinary people feeling compelled to create autonomous AI agents when there will be a litany of companies that will do that for them. Finally, I actually work with AI agents and my only gripe with videos like this is that they do nothing to inspire people as to the potential of AI agents. Nothing demonstrated as ever really cool enough to have the average person who doesn’t really get involved with AI, become very excited. This is low effort, content designed to crank videos out and keep the money train rolling.

[@Leadgen-lu]

Thebfrench voice ...

[@alphamindset9634]

Ching Ching $$