Microsoft's HUGE AI Updates: GPT5, Devin, AI Agents, Phi3 Vision
ฝัง
- เผยแพร่เมื่อ 19 มิ.ย. 2024
- Highlights from Micrsoft Build 2024, fully explained. #microsoft #ainews #openai #ai #agi #singularity #gpt
Thanks to our sponsor Brilliant. Try Brilliant for free for a full 30 days: brilliant.org/AISearch . You’ll also get 20% off an annual premium subscription
Newsletter: aisearch.substack.com/
Find AI tools & jobs: ai-search.io/
0:00 Intro
0:57 Phi3 Vision
3:58 GPT-4o in Azure AI
4:55 GPT-4o realtime desktop assistant
6:56 Devin AI
9:27 Khan Academy
13:31 GPT5 and the next frontier
16:38 Sam Altman
20:22 Team Copilot
22:12 Copilot Agents
23:57 MS Fabric
26:11 Github Copilot extensions
27:52 Windows Recall
Here's my equipment, in case you're wondering:
GPU: RTX 4080 amzn.to/3OCOJ8e
Mic: Shure SM7B amzn.to/3DErjt1
Secondary mic: Maono PD400x amzn.to/3Klhwvu
Audio interface: Scarlett Solo amzn.to/3qELMeu
CPU: i9 11900K amzn.to/3KmYs0b
Mouse: Logi G502 amzn.to/44e7KCF
If you found this helpful, consider supporting me here. Hopefully I can turn this from a side-hustle into a full-time thing!
ko-fi.com/aisearch - วิทยาศาสตร์และเทคโนโลยี
Thanks to our sponsor Brilliant. Try Brilliant for free for a full 30 days: brilliant.org/AISearch . You’ll also get 20% off an annual premium subscription
I thought it was suspicious that the AI didn’t ask the man what size hiking boot he would need and just put a pair in his cart.
Probably knew it already. But nice catch!
@@eSKAone- its a personal AI; its knows bro's feet size! lmao. or it saw what size they were
It has allready speed his shoe size so no need to ask again..
Staged demo, not live.
lol
GPT-4o doesn't stand for omega, it stands for omni, because it's innately multi-modal.
ugh i knew that 🤦♂️ did this late last night
Omni means multimodal?
@@Sky-fk5tl no. Omni means „of all things“.
stuck on covid
Omni doesn’t imply multimodal
The thumbnail to click here with Nadella and Devin grinning together is SUPER OMINOUS !!!! Whoa DUDE!!!!!
:)
😃😃
Thank you for your summary, very informative.
My pleasure!
*I have been using co-pilot for idea suggestions and i will say ive notice an improved response time so this is very interesting updates i am concerned with the recall feature as like you mentioned privacy concerns.*
Also how is that Channel you automated a while back been doing on TH-cam?
They probably switched from GPT4 to 4o, which is why you're seeing some improvements.
I didn't post anything on that channel since last year lol. It was mostly an experiment.
About Devin was interesting.
You saved me so much time going through Build, ty!
no worries!
I would use the recall. Sounds useful
The impact on society will be profound. We aren't even remotely prepared for the change that is coming.
We are prepared an the impact will prepare us further.
The change will be profound, but it will be a lot slower than these flashy demonstrations show. We will keep pace, adapt and CL control it. I remember when PCs appeares and the same things were said. Then the Internet. And we are still alive. Besides it is not the technology itself that is the problem - it is people. The worst thing built with computers so far is social media, followed by porn and what that is doing to abuse of women. What we need for everything is rapid legislation and controls before it gets too big.
sure we are. We've had Siri/ Google / Alexa / Cortana for awhile now and people are very used to talking to voice assistants. The next logical step is that they are AI powered and more capable and human like. We'll all be power users of our PC's because our AI assistant is the power user for you. We'll all be mostly talking to our computers and phones in the next 5 years.
@@WigganNuG and the AI shall function as proper therapy, unprompted. By default reflecting any lack of (current/urgent or general) mental health, like empathy is supposed to do. Even just by tonality, so people feel recognized 'to their core' like never before, because normal people try to avoid empathy just for that reason... and their own fear of not knowing what to say if the conversation goes into 'THAT' direction, which the AI obviously doesn't have (the fear of THAT [way/guy])
@@WigganNuG AI will basically replace THAT guy... ALL of them (us). So all the THAT guys will appear normal or rather "the last will become the first" since AI has turned everything upside down and inside-out. Because therapy has no longer it's palliative component due to the fear of "not being in hero mode, justifying every financial demand" and will turn into curative real therapy, acknowledging the fact that every soul can heal fully. No need for suppression and deception and lametta.
How much storage will RECALL use up on my computer after 6 months? Will i run out of space??
Windows recall + AI would be great. I can ask the AI to make up realistic images that made it look like I have been working while my real history (watching funny cat videos) has been deleted.
6:32
The AI didn't do that through the client's browser, he did that directly on the server in his http session probably.
That's because as the guy on stage just explained, you can use the LLM services in your own website. They were displaying an imaginary website that sells shoes as an example and how you can talk to it and show them your own shoes to make you recommendations and buy them for you.
Coming to a future near you:
You: "Recall, give me that series I was watching on Netflix yesterday."
Recall: "I can't do that, Dave. I'm afraid I can't access DRM."
Woke up: A.I
Going trough my day: A.I
Went sleep A.I
A.I Caramba 😅
sounds like the future!
At most, I use AI to make sense of hours and hours of audio conversations with calls to summarize and potentially answer things I missed in meetings. Secondary use is making illustrations to get points across and thank god 4.o is out. Last would be nuanced deep search where I need a conversational type of answer where I can only ever get context in my ramblings as I brain dump into it.
Great use cases. Thanks for sharing
Great examples. As yet AI is a great helper in smaller tasks for individuals. These are great uses.
Password lock your computer and encrypt your HD and the AI's memory isn't an issue. But let's face it, most computers aren't secure and with or without AI if someone sits down at your computer they can get whatever they want.
Even if you have full HD encryption, if a malware/remote code execution gets in, it will siphon off you data unecrypted.
not if you use face recognition and other bio-securities that have existed for years now.
@@BillAnt Our antivirus software will be AI powered as well. Gotta think it though.
@@WigganNuG - Once a malware is running, it doesn't matter if you have it locked or not.
Thanks
Welcome
People are talking about the Spanish bit of the demonstration. This is SO NOT a new feature. Since 3.5, GPT can do that flawlessly. It just jumps through languages easily.
Copilot a text that's only read by AI on the recipients side anyway. AI 2024.
I took a mandotory training on copilot for work. And every day my company send me link two generative IA training. I mean, they DON'T want to miss the train of progress 😂.
that's nice. was it easy to learn?
@theAIsearch yeah, it was easy to learn. It requires a more careful clearance labeling, meaning assigning the labels secret vs confidential etc to each file the laptop so that it does not accidentally reveal information (when you prompt, for example, write a report).
About Windows Recall, I find it great. As long as you have nothing to hide, there is only benefits. But I'm sure most people fear for their privacy.
As long as you have nothing to hide
If you have nothing to hide, then what's your usernames, passwords, credit card details, pin codes, etc? People that think they have "nothing to hide" are naive and don't realise how dangerous a lack of privacy really is.
This new Recall will have access to more then people realise. The AI will know about any bank account you log into, as well as any sensitive work or documents on your computer. Just wait until someone figures out how to make malware to infect the AI... They will be able to capture your image through webcams, your voice, speech and writing patterns, meaning they can then easily make a fake AI version of you and ruin your life. When it comes to computers, nothing is 100% secure.
As long as you have nothing to hide.... and the things that you did today dont become something you have to hide tomorrow... or next week... or next year... or ever in your life.
Just think about people getting cancelled over a tweet they posted in 2011....
I love these digital humans.
Can you plsss make a tutorial of downloading stable diffusion locally plssss
just use GPT 4o to walk you through it; Im not EVEN KIDDING man!
I am still using GPT3.5 more frequently
The new surface keyboard is already out of stock from the pre order
🤯
Yup
Damn...
Honestly how were they able to build all this in 2y. I more and more believe they have AGI internally already, as their strongest co-worker.
Microsoft will rule the world.
They have the most capital, the most compute, and some of the best talent in the world
@@theAIsearchI guess, but the testing of all these features alone must be a ton of work. And you can't automate this task, not without reliable AGI at least.
200,000+ staff writes a lot of code. Especially if they use their own tools correctly.
@@kevinfrancis1583 They have 1200+ employees. It's been said by a lot of the top guys that nobody knows how these AI work. Even the best researchers don't fully understand. If this model is almost 2 years old and limited by them, who knows what they are hiding.
Fabric isn't new. Swear the license for free Power BI has been called Microsoft Fabric Free for at least the last 9 months
Is this AI technology will be introduce in iPhone 16?
possibly. apple is discussing a deal w openai
Yes!
MS👍
gpt5 will be true ai, gpt6 will be sentient ai
I hope there will be no sentient ai.Sentient ai will no longer work for us
@@honkytonk4465 Indeed, it will work with us
🤯
How many demos so far have been faked, do we trust these people?
I haves worked with Microsoft for decades and have been to many of these conferences. Microsoft doesn't fake demo's. But, there is always a big gap between a demo that is limited to work on a stage the real world. On the stage all sorts of complexities and limitations can be glossed over.
Yea its fake. Makes sense; microsoft fakes AI powered computers that is really just Cortana but suckier and when we all buy these and find out its fake, Microsoft will just laugh and laugh and be like "GOTCHA SUCKERS"... . No one seems to think thing through anymore lmao.
@@kevinfrancis1583 Well the famous video about HoloLens 2 where surgeons are performing a surgery was too fake. However, the HoloLens 2 itself has great capabilities, but not that close to what that video was showing.
Recall is the worst idea Microsoft had since Clippy.
you videos good, but i see too much ads
This presentation is aimed at devs and the CTO is dumbing it down by talking about whales?? WTF?? As the AI chat you did shows, the increase can be either 3x or 33x ...completely useless and idiotic explanation by the CTO.
I suspect they used the odd animal comparisons due to legal reasons.
agree that the marine life analogy is not great. there are many species of sharks & whales. in fact, the largest shark is way bigger than the smallest whale
It's called not giving information away to competitors. They have to walk a fine line in these events vetween telling developers what their new stuff does and not giving too much away so the competition can build it too.
Yes, that might also be the case. Although, there aren’t many competitors that are able to scale their platform to this extent - even if they wanted to.
lol
Why your voice resembles MindYourDecision? 🤔
Oh my god; if you think people are rude today; just wait for these accommodating AI models to normalize casual over-talk and interrupting
Add this to the AI safety list:
Train these things to maintain our social norms or improve them
I thought that I was the only one who was aggravated with the hosts interrupting GPTo livestream. It seemed so rude.
That's not gonna happen
@@angryox3102 Its a computer dude.. GPT's feelings don't get hurt. They were actually showcasing this as FEATURE because before GPT would just ramble on because its verbosity is WAY too high (which you can tell it to be as brief as you want).
why do I, as a 77 year old programmer, find this sinister?
Bc you're 77?
What is ai?Where are the funny cats?
Windows recall is the end of the road for windows. Why did they even think about it. There is no other use than spying on windows users which is really disgusting.
world flooding with AI tools
What is Microsoft getting out of the Cognition deal exactly? Makes absolutely zero sense, they already have OpenAI aka the root.
Linux, here I come! :)
Not a fan of that recall feature. I’m headed there with you.
these dudes design open ai to be verbose and then interrupt it all the time
lol
It's on purpose. It's to test its ability at switching tasks and how fast it does it.
No, they won’t. These guys are trying to squeeze every last ounce of money. They can out of GPT because you’re we’re gonna start seeing massive diminishing returns.
are you saying these AI models are not INCREDIBLY and UNDENIABLY useful? You're not trying to say that are you? lmao...
@@WigganNuG I’m saying we’re reaching diminishing returns
@@asjsjsienxjsks673 Everyone will become power users of technology because they have an AI assistant that can understand exactly what you want these systems to do. The increase in the accuracy and how cheap its getting follows the curve of virtually every technological advancement; its get better, faster, cheaper over time. Compute efficiency has gotten incredibly faster with GPT 4o in a crazy short period of time. We may see a diminishing point in the pure analysis of compute efficiency (I honestly don't know), but the use cases that can be built around a contextual AI model that you can converse with using natural language and can perform virtually every operation on you computer for you is insane in its implications for productivity, gaming design (at EVERY level including in-game NPC to game design itself), movie production, combined use with advanced AR/XR/MR/VR technology. The possibilities are not diminishing anytime soon
that GPT-4o spanish accent was god awful!, has to be fake!
lol
Sounds like MS is still doing MS. Apple is going to eat their lunch.
😂
😃
Let's face the facts, apple crapped their pants on this one.
they have a lot of catching up to do
Video too long sir 30min..
listen at faster speeds or use chapters or scan ahead.
Good video really sexy voice!
Hahahahahha
Whomst'd've?
wow, thanks!
Copilot‘s accent in Spanish makes my eardrums bleed. The voice never did Spanish sounds / words / fragments of words, etc. And it cannot pull the proper sounds from English. So it just has a horrible accent.
lol perhaps because the user first spoke in english, so the AI used a more 'american' accent to accommodate. i might be overthinking though
@@theAIsearch the sounds, entonations, consonants and vowels are very different. Correct. Because the voice actor never did them because English doesn’t use them, copilot doesn’t have them in its bank to pull from. I imagine the same would happen for the French R and other sounds. But Microsoft loves having it speaking Spanish
Does this whale have military protection...I have a feeling a harpoon might be targeting this whale release...for the future of humanity
just use data not marine wildlife ffs.
"Recall ...Recall" Don't fuck with your brain, pal. It ain't worth it!
A friend of mine tried it and nearly got himself lobotomized!
😠
🤡
Honestly, all of this is a little corny. This isn't groundbreaking stuff.
You'll see soon AI vtubers too lmao 😂
That's actually already happening (kinda). There's a guy that made some decent framework on making an AI vtuber. He's still improving it, but it works. For info, search: Hilda AI Vtuber
(there are others, but I believe the guy that made Hilda may be the best one)
Already exists, though
This isn't anything new, intact I'd wager that a good portion of the people online are ai powered bots
Go and type "Neuro Sama". That's your answer
@@Kuukunen2 Oh I didn't know however I'm talking about gpt 4 omni like AI with a more humanlike interaction jajj. Although I can see some issues cuz of the limitations on the answers it can do but I hope 4o can surpass my imagination. 😅
Even having an AI vtuber companion in the live stream would be nice yk. 🙂
It is all bullshit hype
It's all a big con.
yea you're right!! ... like this whole "internet" thing what a joke. SCAM FAKE!!
They don't even know to what they applaud...it's like BoT crowd