Transcription
Here's the AI news you probably missed from this week. There was a ton of news out of OpenAI this week. Let's start with the GPT 5.5 Instant. This is OpenAI's brand new default model inside of ChatGPT. Now, if I'm being totally honest, I don't feel like most people are going to notice a massive difference with this. If you're using ChatGPT, you'll probably notice some fairly minor improvements. They claim it's smarter and more accurate with clearer, more concise answers that feel better tailored to you. This isn't like a brand new top-of-the-line model out of OpenAI. This is more of like an incremental improvement to the previous default model.
So, here's some side-by-side comparisons that they share on their website here. This is the old default model, GPT 5.3 Instant. This is the new default model, GPT 5.5 Instant. They give it this math problem here, and they ask if it looks okay. We can see in the old model there's a lot more explanation that goes along with their response here. As it goes through the response, it actually ends up finding no real solution. So it actually starts off by saying yes, this is clean and correct. Sort of thinks through it some more and then at the end says, "Wait a second, no, there's no real solution." Compared to GPT 5.5 instant. Again, we can see it's a little bit more concise as it walks through the math. There's less explaining as it goes through the math. But when it gets down to the bottom here, it actually does come to a solution. Since x is greater than or equal to one, the valid solution is this. Where the first one couldn't find a valid solution.
Here's another example that shows that it's just more concise. You can even just see in the size of the scroll bar here that GPT 5.5 instant, you have to scroll less. "How do I tell my coworker to quit yapping all the time?" And we can see it gave a long sort of more detailed answer where this version here gave a much more concise answer. Here's another example of writing feedback. Once again, you can tell by the scroll bar that this one's a lot more detailed of an answer where this one just gets to the point quicker.
It also claims to be more personalized. So, we can see in the previous model, "What new boba should I try?" And it gives a fairly generic answer. "If you want something that feels more new SFT scene and less standard boba chain, I'd narrow it to these." The GPT 5.5 Instant. "Since you already hit Asha Tea House pretty often and seem to prefer cleaner Taiwanese high mountain tea vibes over sugary boba, I'd push you towards one of these depending on the mood." So, a little bit more tailored to what the model knows about you from its memory.
This model is available to all ChatGPT users. That means if you're on any of the paid plans or the free plan, you're getting this model as your new default model. As a quick side note, I figured I'd add this news in as well cuz it's relevant. The new GPT 5.5 Instant is also available inside of Microsoft 365 Copilot. So if you do use Microsoft's C-Pilot platform, you're also getting this model inside of there.
Now, jumping quickly into ChatGPT here, the model selector has recently relocated. It used to be up in the top left. If you click over here where it says "thinking" on mine, it might say something else on your default right now, but if you click here, you can see the latest is 5.5. And you have the option for instant thinking or pro depending on which plan you have. You may not have pro. You can click configure and change the model. Right now I'm defaulting to 5.5. And I also have mine defaulting to thinking. But for most people you can default it to instant and have it auto switch to thinking if it needs more thinking. Again, this isn't a new state-of-the-art crazy new model that's like blowing everything out of the water on benchmarks. It appears to be a new sort of refined version of the instant model that we used to get as the default.
OpenAI also showed off a brand new voice mode that looks pretty dang impressive. So this week they rolled out some new real-time voice models. There's actually three new models to be exact. There's GPT Realtime 2, which is a voice model that uses GPT 5 class reasoning and can handle harder requests. There's GPT Realtime Translate, which is a new live translation model that translates speech from over 70 input languages into 13 output languages while actually keeping pace with the speaker. And then GPT Realtime Whisper, which is a streaming speech-to-text model that transcribes speech live as the speaker talks.
Now, these models are actually really, really impressive when you see the demos. Unfortunately, they're only available in the API right now. So, if you're expecting to see these in ChatGPT or inside of Codex, well, as of this recording, they're not there yet. You can, however, get a limited demo here of GPT Realtime 2, which is the sort of chat back and forth audio. But before we test this, check out some of the highlights of this demo here. So, the first demo is the translation demo. What's really impressive is that the model can listen to me and translate while I'm speaking. It waits for the keyword, like the verb, to start translating right away and the result is a much more natural conversation just like a dialogue between two people. I can even interrupt in German and the model switches effortlessly between my German and your French and we can even include technical terms like GPT, real time, OpenAI or computer use and the model has no trouble handling that.
>> And then the next demo he shows off here is the real time 2 demo. The one that you actually can play with to a limited degree right here on their website.
>> Hi there.
>> Hi again. What's up?
>> Yeah, I have uh a customer meeting coming up. Can you take a look at my calendar?
>> You have a meeting with Sable Crust Robotics in 12 minutes and you're meeting with Alex Kim, their CTO.
>> Great. Thank you. Oh, please stay quiet for a second until I say back to demo.
>> Ramon, don't forget now that these models have things like reasoning and parallel tool calling, it's even more important to use things like preamble. This way, the model can explain itself and update the user.
>> Thank you, Jason, for the great reminder. Very important. Actions can of course take a few seconds and so it's very important for the model to acknowledge those. With GPT real time too, you can communicate directly during the reasoning and the tool calling so the user stays informed. And by the way, what makes voice agent so natural now is that they stay in the conversation. Jason and I have been chatting. Uh the model has been listening and it's still listening now but not interrupting us until I say back to demo.
>> I'm here when you're ready to continue the demo. Could you update the CRM and put the meeting of today as a brief and the next steps?
>> Let me pull the latest context and update your CRM. Sablerest launched warehouse automation this morning. Expansion is active. Security review is the blocker.
>> All right. Thank you.
So for me, out of all of that, I think it's cool that it can connect to a lot of the tools and sort of sync to things and start to feel a lot more like this personal assistant agent kind of thing. But the thing I was most impressed by was you can kind of tell it to shut up for a minute while you have another conversation, which is something I know all too well because I'll be trying to demo stuff and I need to like mute my microphone so it doesn't hear me. But now I can just tell it like, "Hey, be quiet. Just listen in for a minute." Let's go ahead and start a session and play with it real quick.
"Tell me about the MattWolf YouTube channel."
>> The MattWolf YouTube channel is known for covering AI tools, tech trends, and practical workflows, often aimed at helping creators and professionals use new software more effectively. In this demo, I can't browse for the latest updates, but the channel's focus is generally on exploring emerging tech in an accessible tutorial style way.
>> Okay, just be quiet for a second uh while I say some other stuff. I'll let you know when you can jump back in.
>> Got it. I'll stay quiet until you tell me to jump back in.
>> So, yeah, it's interesting that it finds my channel to be very tutorial-based when obviously it's mostly news-based. Lately, I've been doing a lot more tutorials just because I've been finding things that are uh really fun for me to teach and play with, and I want to show off the cool tools that I've been playing with. But really, the focus has been news. So, it's interesting that it finds that. Okay, you can jump back in now. I'd love to hear your thoughts on what I just said about the channel.
>> Nice. Let me think about how your description of the channel fits with that earlier impression. That makes sense. If most videos are news-driven, tutorials can still look like the main focus because they're concrete and easy to categorize. To steer perception, you could label series clearly, like AI news versus hands-on. Use playlist.
>> And I got cut off. We only have a minute and a half. So, you get the idea, though. I really wanted to test pausing it, talking for a little bit, and then jumping back in. Was my discussion a little self-serving? Maybe a little a little bit of a bummer that it's not in ChatGPT yet and it's not in the Codex app yet. But from everything I can tell, it will be very soon. And if you do want to use this, you can actually use it inside of the OpenAI playground. So if you go to platform.openai.com/audio/realtime, you can use this model here. Although this is connected to your API, so it will use API credits. There will be a cost to use this, but I imagine we'll see it inside of ChatGPT and the other platforms really soon.
Another new feature inside of ChatGPT this week is the trusted contact in ChatGPT. So this is designed so that you can add a trusted contact inside of a ChatGPT account and then if you start having chats about distress or harming yourself or things like that ChatGPT can actually contact your trusted contact and let them know that these sort of discussions are happening. It's an optional safety feature in ChatGPT that allows adults to nominate someone they trust, such as a friend, family member, or caregiver, who must be notified if our automated system and trained reviewers detect the enrolled person may have discussed harming themselves in a way that indicates a serious safety concern.
Next up is this new tool from HubSpot called AEO, which stands for Answer Engine Optimization. It's basically a dashboard for marketers where you can plug in your brand and product and it will reveal how you show up across AI search. So it'll show your visibility in tools like ChatGPT, Gemini, and Perplexity. What prompts triggered your brand as a recommendation or output? How you compared to other competitors and what to do to improve. I tried it with my newsletter Future Tools and it actually showed me that in terms of AI outputs, Future Tools is kind of getting buried. Like AI seemed to recommend other AI newsletters, but much more than mine in response to a variety of prompts, which is something I definitely need to work on. Thankfully, HubSpot AEO will also give me a clear plan for improvement. So, I'm really glad I found all of this out now rather than later. And this is actually really important right now because more people are using AI to discover products and services. And those answers are directly influencing which brands get seen, trusted, and chosen. I mean, if you're watching this YouTube channel, you're probably like me, and the first place you go when you need a product recommendation is some AI tool. So, if a brand hasn't checked out how they're showing up there yet, they're missing out on a big chunk of user data. If you want to see what AI is saying about your brand, try HubSpot AEO free for 28 days at the link in the description box. And thank you so much to HubSpot for supporting my channel and sponsoring this portion of today's video.
OpenAI has also cooked up some new features inside of Codex this week. Like now Codex works directly in Chrome on Mac OS and Windows. I've been personally using Codex a lot more. It's kind of become my go-to platform for writing code and even doing a lot of my chatting. Honestly, I built a whole second brain in here and I even made a whole video about how this second brain works. I highly highly recommend checking out this video if you haven't already. It's uh one of my favorite videos I've made in a long time because it really shows a good real-world use case for AI where I save a whole bunch of YouTube videos into this system, a whole bunch of articles, things like that, and then I can chat with it and instantly retrieve anything I've saved. It's like this really cool second brain thing. And I think you'll find that video really helpful if you haven't watched it already. It's a really cool process.
Anyway, back to Codex. If you come up to plugins in your Codex app, you should now see the option to install a Chrome plugin. So, I'll go ahead and hit plus on this. I'll click install Chrome. It's going to ask me to install the extension here directly in Chrome. So, I'll add that to Chrome as well. And we can see Codex is now officially connected to Chrome. So, if I snap my Codex window over to the left here, and then my Chrome browser over to the right here, I can start a new chat here inside of Codex and say, "at Chrome, look at futuretools.io/news and give me a breakdown of the news that came out today." It says, "I'll use the Chrome plugin for this since you explicitly pointed me there. Then I'll filter the future tools news for today's date. Chrome skill loaded." Okay, so it opened a new tab with the future tools news page. We can see that it sort of highlighted the tabs. So it's working in those tabs. Then it eventually came back with this, "Chrome connected but direct page automation was blocked by an open extension UI." I don't know what extension UI it is open, but okay. It did eventually give me everything, but it did it through a search. So, the Chrome plugin in Codex is working. We saw it sort of take control of Chrome, but it's still a little buggy from my own testing. Again, the day I'm recording this is the day that it was released, so I imagine it's going to get better really quickly.
But that wasn't the only feature we got out of Codex this week. We got one that is a little random. OpenAI released pets inside of Codex, which is just like a new little character that sort of lives inside of your Codex. You can see in their demo it's this little like cloud robot character that they spawned in. It's funny because we're getting all this news from OpenAI about how they're eliminating side quests and then we get like these little pets that live inside of Codex. So if you use `/pet`, it'll wake your pet. If you use `/hatch`, you can actually customize what you want your pet to look like. I've seen some pretty interesting pets already. I think my favorite pet that I've come across is this one from Grz or Gru, which is the angry Dario pet, which is just like a mad Dario that runs around. And if you yourself want an angry little Dario running around in Codex, well, there is a platform available over at codex-pets.net where you can see pets that other people have created and then pull those pets into your own Codex platform. So if you want like this little wave or if you want Alan Turing or if you want the Doom guy, you can pull these pets directly into your own Codex on this website. Let's search for Dario because I'm pretty sure somebody put angry Dario in here. Yeah, you've got yelling Dario and you've got this same angry Dario that we were just looking at. If you want to use this Dario, you click in here. It gives you a little command that you can run in your terminal to pull in angry Dario. Here's the various animation states for Dario. But let's go try it out real quick. This is just silly and not super useful to anybody, but you can tell it's something that the people on the Codex team had some fun with.
So, let me pop open Codex here again. I'm going to start a new chat. And if I just type `/`, it woke up this little robot guy here that's now my little pet that can hang out inside of my Codex with me. So, I'm going to go ahead and close my pet here. If you come up here to plugins on the top left, come up to skills, type hatch, you can see there's a recommended skill called hatch pet. Let's add this skill. And now this skill is installed. Now, if I go to new chat and I type `/hatch`, we have our hatch pet. And let's do a cartoon wolf character. So, it's getting Wolfie ready, imagining Wolfie's main look, picturing Wolfie's pose, and then hatching Wolfie. So, we've got Wolfie's main look here. So, this took a lot longer than I thought it was going to take. Almost a half hour, 30 minutes here. But now, let's see. If I type `/pet`, does it hatch Wolfie? No, it still brings this dude. Okay, so apparently when you hatch a pet, it doesn't immediately work. It says open settings, go to personalization pets. Under custom pets, click refresh. If Wolfie is not visible yet, select Wolfie, click wake pet. All right, so I go to settings here. Go to personalization. So, Codex gave me wrong information. If I go to appearance, I see pets down here. And my custom pet Wolfie is right here. So now I can select that. Now if I go and type `/pet`. Okay, now it woke up my wolfy pet here and he runs around. Okay, cool. So that's how I generated my little wolfy pet. Not the most intuitive process, but at least now we know how to do it. Oh, interesting. Even when I'm outside of Codex, he's still there. Maybe I can get him to just like rest on my head there.
There were a few announcements out of Anthropic this week, like this new dreaming outcomes and multi-agent orchestration inside of Claude managed agents. So to me, this feels like Claude trying to become more and more of what OpenClaw and Hermes and some of these other sort of proactive tools are. So dreaming extends memory by reviewing past sessions to find patterns and help agents self-improve. We're also making outcomes, multi-agent orchestration, and webhooks available to developers building with managed agents. The most interesting part of all of this to me is specifically the dreaming. Dreaming is a scheduled process that reviews your agent sessions and memory stores, extracts patterns, and curates memories so your agents improve over time. Dreaming surfaces patterns that a single agent can't see on its own, including recurring mistakes, workflows that agents converge on, and preferences shared across a team. It also restructures memory so it stays high signal as it evolves. So to me, this doesn't seem like normal memory like what we've already got in ChatGPT and Claude where it just kind of remembers everything that you say over time, kind of keeps key points of that and then sort of dumps it into the context every time you talk to it. This is more of like a proactive sort of behind the scenes constantly looking at and reorganizing memories and trying to find patterns and better understand stuff on like a regular schedule. Now, this doesn't appear to be something that's just inside of the standard Claude or inside of Co-work yet. It's specifically inside of the managed agents, but I do kind of feel like this is all headed in a direction where something like this gets baked into Claude Co-work so that Claude Co-work can eventually become closer and closer to what like OpenClaw and Hermes are where there's just these scheduled tasks running behind the scenes constantly that look at the memory and try to improve the system based on what it knows and how you've used the platform.
In fact, this is where I think most AI is going. That is sort of like the big theme that I'm seeing is these AI agents becoming more and more proactive as opposed to just chatbots where you go to them and you prompt it with a question when you have a question and get a response like a sort of glorified better version of like a Google search. But they're going to become more and more proactive where it starts to find patterns and then reaches out to you and says, "I noticed this is happening over and over again. Would you like me to resolve that so it stops happening? I've noticed this pattern in your workflow on how you use Co-work. Would you like me to make that automatic instead of you having to prompt me all the time?" I think we're moving into this world where these AI agents are going to get more and more and more proactive where it notices what you're doing and then comes to you and says, "Hey, based on how you've been using me, here's what we can implement to make your life easier." At least that's where I hope it's going cuz I think that's like a really cool use of AI. But we'll see how it all plays out. This dreaming feature feels like a step in that direction for Claude.
Now, the other big news that came out of Anthropic this week was a sort of two-part announcement. One that most Claude users are going to be pretty happy with. They're actually raising the usage limits for Claude. But the other one is one that I don't think anybody really saw coming, and that's that Claude signed a compute deal with SpaceX. "We agreed to a partnership with SpaceX that will substantially increase our compute capacity. This along with our recent compute deals means that we've been able to increase our usage limits for Claude code and the Claude API." That's like the main reason we're hearing most people going back to ChatGPT after trying out Claude is that I only use it for a little while and then I hit usage limits and need to wait until 2 a.m. before it refreshes again or whatever. So now they're lifting some of those limits so you can use it a lot more. And they're doing that by signing deals with SpaceX for compute. And then the other deal they're talking about is that they committed to spending $200 billion on Google's cloud. So they're getting compute from Google and they're getting compute from SpaceX. And one of the reasons this feels so interesting is cuz well, Elon Musk hasn't really been a fan of Anthropic. Calling Anthropic "misanthropic" only, you know, two months ago here. "As recently as March, is there any more hypocritical company than Anthropic? Last year saying winning was never in the set of possible outcomes for Anthropic." But fast forward to today and Elon and Dario seem to be making deals together. A lot of people are speculating this is kind of a "the enemy of my enemy is my friend" situation. Obviously, Elon is in a lawsuit with Sam. Sam and Elon really aren't getting along. And so if Elon can then go and help Dario and Anthropic get further ahead than OpenAI, then I think Elon sees that as a win over OpenAI. And this all kind of makes sense as well when you think about it because there's the XAI model, Grok, which is a decent model, but it's not like a state-of-the-art best of the best model. Those belong to OpenAI, Anthropic, and Google. Most people don't really put Grok in the mix of the best models right now. Elon's also said something like the amount of compute they have, they're only using like 11% of it. So, they have all of this excess compute. Most people aren't really going to the Grok models as much as the Anthropic and OpenAI and Google models. So, might as well make some money off of all of this excess compute that is not getting used by XAI, which is now, I guess, getting rebranded as SpaceX AAI. True story. Look it up. But they have a lot of extra compute is the point. And they have no problem selling it to Anthropic if it means that Anthropic could potentially pull ahead of OpenAI again.
Now, speaking of Grok, at the end of last week, we did get another model out of XAI. XAI launched Grok 4.3. If we look at this post from Artificial Analysis here, we can see how Grok scores. Now, Artificial Analysis sort of takes a whole bunch of benchmarks and kind of looks at like what is the smartest model sort of across the board across a whole bunch of benchmarks. And we can see that the old Grok model was way down here. This new Grok 4.3 did make a pretty big leap, but like I said, nobody's really seeing it as like the most state-of-the-art best of the best models. Those go to OpenAI, Anthropic, Google, and then we even have a second OpenAI model, a Kimmy model, and a MiMo model. All still ahead of this new Grok 4.3 model. Where this one really excels though is how inexpensive it is to use. If we look at Claude Opus, that's the most expensive model and you know another Anthropic model and three of the OpenAI models all fall in that same super expensive category where Grok 420 was way down here. The new Grok 4.3 jumped to even cheaper. So it's a good model that sort of is, you know, fairly in line yet not quite as good as the best of the best models, but it's also way cheaper to use.
Moving along, another week and a lot more AI drama to talk about. So, let's dig into it. The drama this week once again revolves around the trial between Sam Altman and Elon Musk. All sorts of information that's just absolutely fascinating is coming out during this trial. For example, we learned that OpenAI co-founder Greg Brockman's stake in OpenAI is worth $30 billion. If we remember, the whole lawsuit is about Elon Musk suing OpenAI, saying that it was supposed to be a nonprofit not to enrich the founders. And well, his claim is that the fact that Brockman's stake is worth $30 billion feels pretty enriching towards the founders. We also learned that there was this text exchange between Greg Brockman and Elon Musk. Elon said, "Would you be open to a settlement?" Greg said, "We should both just drop our claims and end this." Elon responded, "By the end of this week, you and Sam will be the most hated men in America. If you insist, so it will be."
But probably the biggest story to come out of this week and this trial is around what happened when Sam Altman got fired a few years ago and then eventually rehired. Mira Murati, who is the former CTO of OpenAI and the interim CEO for like a day when Sam Altman was fired, testified under oath that Sam Altman lied to her about the safety standards for a new AI model. Murati said Altman falsely stated that OpenAI's legal department determined a new AI model did not need to go through the company's deployment safety board. Marotti testified that after speaking with Altman, she checked with Jason Quan, its general counsel, and then said that that's where she realized there was a misalignment between Quan and Altman. "I confirmed that what Jason was saying and what Sam was saying were not the same thing." She said, "To be safe, she made sure that the model went through the safety board anyway, even though Sam said, 'Don't worry about it. Legal cleared it.' And then when she asked legal, legal said, 'No, we didn't.'"
But the absolute wildest part of what we've seen so far are the text messages between Mira Murati and Sam Altman while all of this was happening on the day that Sam was being fired. This text thread here is just kind of crazy to me.
Sam asks, "Can you indicate directionally good or bad? Satya and others are anxious."
Directionally very bad. Okay. Can you wrap up soon? Lots of pressure from Microsoft for an update.
Sam, this is very bad. Can I come in?
They don't want you to. What do you want to make it better?
I'm still willing to just walk away if that helps.
If they are ramped up for crazy lawsuits against me, then I'm not sure what.
Can you please tell them I just want to resolve this, however, and would like to join?
They're convinced about their decision. For me to be fired or some new thing?
Yes. For you to be gone?
Okay. Then can I come in and talk about a path forward with them?
They're saying no and they need more time.
More time for what?
They've walked me through all the reasons and the issues with you and why you can't be CEO.
Can you ask why they've been saying all weekend they wanted me back?
They want a new CEO in place.
Can you say you will call back in 10 minutes?
They want to have a new CEO in place tonight, not me.
Do they know who?
Can I tell Satya?
Is this final or should you add Satya in?
Trying to add Satya now.
Still don't want me.
New guy is rando Twitch guy.
They don't want you, Emit.
Yeah, but hold on. I'm pulling Satya now.
Okay. Do you think any way you can turn this around even if we let them sleep on it tonight or whatever?
Have been trying. Yeah, hoping Satya can help undo this.
Should team send letter to board now?
Is he on with them?
It won't matter. They don't care if everyone quits.
Is that what they want?
Is the IP going to Anthropic?
That's what team thinks. Just not your hand in AGI.
I cannot come back.
They were the ones that asked as of yesterday morning.
Did Satya get on the call?
Yes, with him.
Wait, I have an interesting idea.
Still with Satya.
Go ahead.
What if Microsoft acquires OpenAI? Would that provide the governance the board wants?
Is Satya making progress at all?
Satya's being diplomatic.
Will you emit on with the team ASAP?
Also, does Adam know you rehired me?
Yes.
Just a wild chain of text messages. Like, this is what was going down on that weekend when Sam was fired and there was just all this confusion and nobody knew what was really going on. And to actually see these text messages kind of shows like where Mira and Sam's head was at. And also kind of proves that Satya was like the adult in the room. Like they wanted Satya to get in and sort of intervene and make this all right. And from what I can tell, that's kind of how it went down. Satya was like the adult who came in and said, "All right, everybody just cool down. Let's figure this out." I don't know if that's what actually played out or not, but that's kind of what the text messages feel like they imply.
Okay, I've covered a lot already, but there's a handful more things I want to share, but I'm just going to go over them quickly. So, let's do a rapid fire. Jumping back to Anthropic for a minute. You can now use Claude directly in Excel, PowerPoint, Word, and Outlook, and it will actually remember conversations across platforms. It says, "As Claude moves between tasks in your Microsoft apps, it keeps the full context of your conversation." So you can be having an email in Outlook, jump over to Word, have it reference what you were just talking about in that email in Outlook, and then jump into Excel, and it will remember what you were talking about in Outlook and Word, and all of the context remains between the various conversations in the different Microsoft apps. That seems pretty handy.
We got some updates out of Adobe this week. Now, inside of Adobe Acrobat, you can do a bunch of cool AI stuff. You can chat with your PDFs and you can quickly create presentations, podcasts, blogs, and social posts from the documents. So, it's almost like NotebookLM but built directly inside of Adobe it seems like. So, if I pop open my Adobe Acrobat here, let me open up the "Attention Is All You Need" PDF. It even says, "This appears to be a long document. Save time by reading a summary using AI assistant." So, I can view the summary. It gives me an overview here of the document and then I can chat with the document like, "Explain self-attention benefits," etc. If I click on "All Tools" over here, we've got "Generate Presentation," "Generate a Podcast," "Translate this PDF." Out of curiosity, let's see how the podcast does. Let's do a quick highlights podcast. All right, so our podcast is done. It took a couple minutes, but let's go ahead and listen to a quick sample.
>> Welcome to this new episode. Today, we're diving into a really fascinating breakthrough in AI language processing, the transformer model.
>> Yeah, it's a game-changer. You know, for a long time, computers struggled with understanding and translating language because they processed sentences word by word, step by step.
>> Okay, it's not too bad. I think the NotebookLM voices sound more realistic and natural. But, you know, same kind of idea as NotebookLM, but you can do it directly inside of Adobe Acrobat.
Now, speaking of podcasts, Spotify rolled out a new feature. You can now save your personal podcast to Spotify and listen anywhere. And what it means by personal podcast is, well, AI-generated stuff. It says, "Your agent can generate a daily briefing private to you and it's saved alongside everything else in your library." So if you already listen to Spotify podcasts and music and things like that, you can tell like your OpenClaw or your Hermes agent or or Claude Code or whatever to generate a podcast for you every morning. And because Spotify has its own CLI now, you can actually have it automatically push those podcasts into Spotify and listen to your personalized podcasts along with the other podcasts you might normally listen to. That's actually pretty cool. I mean, I don't have a commute, so I don't listen to as many podcasts as I used to. But it would be kind of cool to have like a morning AI news briefing generated automatically with my OpenClaw and then pushed into Spotify and then when I jump in my car, it's just there and ready. That's that's kind of sweet.
Apparently, they're working on some new AirPods that actually have vision capabilities. "Apple has reached the latest stages of development for new AirPods with built-in cameras, a wearable device designed for the artificial intelligence era. The earbuds feature cameras that act as eyes for the Siri Digital Assistant, allowing the device to capture visual information in low resolution and provide information to the user." Now, I don't believe there's a timeline for these yet, but from my understanding, they're not just like recording everything all the time. You can't take pictures from your ears. It just gives them the ability to see what's around them and use that information as context when you're chatting with AI. But, it's not going to be like the Meta Ray-Ban shades where you can actually snap a photo with them or record video of what's going on at all times. It's just for like in-the-moment context. Again, this is all early and mostly sort of rumors, so I don't know what I'm talking about here, but that's just my understanding from what I've read so far.
Here's an interesting story. Nvidia and PTE Group are partnering with startup SPAN to install mini data centers on the walls of new homes. Each unit packs 16 Nvidia Blackwell GPUs, four AMD Epic CPUs, and 3 TB of RAM and taps unused home electrical capacity to run AI inference workloads. So like new homes are going to have AI built into them. It's going to be like internet or cable. You're just going to have like an AI module built onto your house that's like an AI supercomputer and you won't need to use ChatGPT or Anthropic or places like that. I can also see this playing out in a way where everybody has these like mini data centers on their home and then you like rent them to the AI companies and they pay you to be able to use them. Almost like Bitcoin mining or like if you have solar panels and you're putting energy back on the grid. Well, the energy companies used to pay you for putting energy on the grid. Most of them don't anymore. It could almost be like that type of scenario where you have your own personal compute at home on your own personal data center to run AI whenever you want and then your excess capacity could go to like the Anthropics and OpenAI's of the world and when your compute is being used you earn money for the compute that you generated similar again to like blockchain mining. I don't know which direction it's going to go. I'm inferring a lot here. Interesting concept, but that's what I got for you today.
There is so much happening in the world of AI every single week. It's so easy to get overwhelmed and feel like you're falling behind. And so that's why I make these videos. I try to sift through all the noise and all of the hype and all of the big announcements and bring to you just the stuff that I think the most people are going to be interested in, the stuff that seems to have the biggest impact on the most people. Like I like to say, I'll drink from the fire hose and then try to just give you what you need to know every week. If that's something that sounds interesting to you, maybe like this video and consider subscribing to this channel. I'll make sure more stuff like it shows up in your feed. I've also been doing a lot of really cool AI tutorials lately, breaking down some of the ways I'm using AI in my own life and business. I recently made a whole video about the ultimate AI knowledge base system where you put all of your articles and videos into this knowledge base and then you can easily find them again by chatting with them or journaling. I also built an AI CRM. It's a really, really cool process that I was really excited to share. If you want to see that video, I'll make sure it's on the very end screen of this video. So, if you wait till the end, you'll see the link. I'll also link it up in the description. I think it's a really helpful video, but I'm probably a little biased. Again, I record these on Thursdays, publish them on Friday. So, if any news came out late Thursday afternoon or evening or during Friday, it'll make it into next week's video. But again, that's what I got for you. Hopefully, you learned something. Hopefully, you feel a little bit more tapped in on what's going on with the AI world and you're not feeling so overwhelmed. That's my goal here and hopefully I accomplished that. Really, really appreciate you. Thanks for nerding out. Hopefully, I'll see you in the next one. Bye-bye.
Mostest thing is to see your rea. All right, thanks for waiting.