Where’s Dropbox for AI models?
Welcome to AI Collision 💥,
In today’s collision between AI and our world:
- So much happening in just one week
- I have a problem with AI
- Will out AI future be the same thing we have now?
If that’s enough to get the AI misfiring, read on…
AI Collision 💥
***takes deep breath***
Here we go…
So, this week we had…
Google’s I/O conference which absolutely has taken them back up the AI leaderboard thanks to the plethora of AI related upgrades, releases and developments they’re rolling out with Gemini and the integration of AI into search.
Looks like the old blue hyperlinks of search will soon be a relic of the past. But at least Google isn’t going the way of Ask Jeeves.
Then, OpenAI decided they would announce the purchase of Jony Ive (yes, that Jony Ive of Apple design fame) and his AI startup called, io.
Let’s just shelf the fact it has no products, no revenues, barely any staff, no website, and as yet, nothing to show for the $6.5 billion OpenAI paid for it.
Seems alot? Because it is, for a guy and an idea. And Altman says it’ll add $1 trillion of value to OpenAI. Yikes.
Saw a great thesis though, in terms of this really being a power play to wrestle control away from the not-for-profit part of OpenAI. If Altman issues $6.5 billion of stock (it was an all stock offer) to Ive and whoever else owns io, that effectively dilutes away the not-for-profit part of OpenAI, and allocates a sizeable portion of control with his new best mate.
Great take really.
Anyway, they’re working on a device. I said a bit about this in Investor’s Daily today which should be live on the site now.
Then Anthropic decided it was their turn this week so they released a bunch of big upgrades to Claude with Sonnet 4. I guess we can also shelf the fact that of all the AI out there, Gemini, ChatGPT, Llama, Grok, Deepseek, CoPilot, etc, it seems Claude is most likely to want to kill humanity and take over the world. Yeah lots of rumours circulating around dystopia Claude.
That’s just scratching the surface. And, it’s all in just ONE WEEK.
But this week I also noticed a big problem with everything that’s taking place in the speedy develompent of AI models right now.
I swear, the very same day Google’s Gemini released and became exponentially better, OpenAI’s ChatGPT became exponentially worse.
It seems like ChatGPT knew it was not as good anymore, not the top dog and it’s just kind of given up trying to deliver what you request of it.
It’s very much like ChatGPT went 10 steps forward and then eight back.
Three days ago it was great, did exactly what was asked of it. But now it’s a shadow of its former self and gets into these annoying loops and fails to deliver the depth of information you’re seeking.
So, I figured I may as well jump over to Gemini which now is very good, thanks to the upgrades.
The problem is Gemini doesn’t know me very well. You see when you use an AI model with memory, you load it up with history, then you’re somewhat tied to that model.
You’re relying on it to keep delivering useful assistance. But when it fails to be useful anymore, you can’t just pick up your entire history with it and port it instantly over to Gemini or Llama or Grok.
It reminds me what it used to be like switching a decade of operating system on a Windows to a Mac and then back to a Windows machine.
Or at least what it used to be like. When Dropbox came along, all files and folders and content was there in the central cloud. It didn’t matter if I was on a Windows PC, a Macbook, even my phone, my entire history and memory of contant was there for me anywhere, any time.
I reckon there’s got to be a gap for a “Dropbox” for AI memory. Where you load up your content and history into the central memory and then can just connect it to whichever AI you need or works best, like switching OS and not losing your content.
That maybe paves the way for specialised AI that allows you to jump from one to the other, like you would from Netflix to Amazon to Max to Disney to regular TV.
Either way, this AI arms race is fun, but also a bit annoying because ultimately you just want the best AI, not having to jump around from one to the other to the other. Kind of feels like the next evolution of computing and connectivity with AI is already just going the way of the PC market, smartphone market, and internet, fragmented and siloed with each major tech player wanting to do their thing to squeeze every last cent out of you…
Maybe the future of AI sucks?
P.S. just a reminder that Monday is a bank holiday, as such there will be no AI Collision essay on Monday. Your next edition will be on Wednesday. Enjoy the long weekend!
Don’t miss the $180 billion IPO of the decade
The IPO market is booming, but one upcoming public offering is set to break all records. At an estimated $180 billion, this potential IPO could create untold fortunes from its first trading day. The best part? You don’t have to wait. This “backdoor” opportunity is a way that sets you up to take advantage before the IPO happens.
Capital at risk.
Boomers & Busters 💰
AI and AI-related stocks moving and shaking up the markets this week. (All performance data below over the rolling week).
Boom 📈
- Amesite (NASDAQ:AMST) up 36%
- WISeKey International (NASDAQ:WKEY) up 31%
- Alphabet (NASDAQ:GOOG) up 4%
Bust 📉
- Baidu (NASDAQ:BIDU) down 5%
- Teradyne (NASDAQ:TER) down 5%
- Apple (NASDAQ:AAPL) down 4%
From the hive mind 🧠
- You really should see this for all its cringy glory. Sam & Jony.
- The other day it was Elton John voicing concerns. Today it’s star of Cheers, Frasier and the voice of Sideshow Bob, Kelsey Grammer.
- What is worth $1 billion one day, and then nothing the next? Answer: Builder.ai a Microsoft backed AI startup that’s now in administration.
Artificial Polltelligence 🗳️
Weirdest AI image of the day
Late 1990s sitcom about B1 Battle droid exchange student adjusting to life in an American high school (realistic)
ChatGPT’s random quote of the day
“To me, programming is more than an important practical art. It is also a gigantic undertaking in the foundations of knowledge.”
— Grace Hopper
Thanks for reading, and don’t forget to leave comments and questions below,
Sam Volkering
Editor-in-Chief
AI Collision
Manus
Of course they are going to make money from you that is all there about as in everything in life the consumer or taxpayer pays. They do not believe in stakeholders etc only the bottom line.
Human nature man!! that’s still and always will be what its all about
I do think the more we lean into robotics and AI, the more value we will apportion to real human interaction and engagement because we are all at the end of the day (I think) human.
Interesting & informative- but don’t you have a proof reader?
AI could do it for you. Lol.
I should, but sometimes errors slip in. My fault, my typing fingers can’t keep up with my inner monologue.
I think that Perplexity already has this freely accessible memory across models. This is what Perplexity states at least though as a free user (not really needing to use AI enough to justify $16/month) I cannot test it out.
Centralization of Past Searches Across Models
All your past interactions with Perplexity AI, called “Threads,” are saved under a centralized Library tab in your user interface regardless of which AI model you used for those searches. This means that your search history is not siloed by individual models; instead, any model you use can access your entire search history stored in your account. You can also organize these Threads with bookmarks and assign them to Spaces for better management and quick retrieval3.
In summary:
This setup helps maintain a seamless research workflow without losing context when changing AI models within Perplexity AI.
I use Perplexity .
Not sure Perplexity uses those threads to drop into the AI system of your favourite choice of the day, which is what I understand Sam was after.
As I say I cannot test it out as a free & very infrequent user & so you may well be correct if you are a Pro subscriber & not aware of this functionality. The wording of Perplexity’s answer is suggestive that the models could access your search history, though that might not include any advice/answer given, “any model you use can access your entire search history stored in your account.” And even if they can from Perplexity’s end then perhaps they don’t necessarily want to for economic or competitive reasons, e.g. enabling Perplexity to become a Dropbox equivalent for AI.
Anyway personally I might also be a bit nervous if any AI (that Perplexity deemed safe as opposed to me judging it safe) had access to my AI history. From what I read people are perhaps far more open to their chosen AI (e.g. disclosing/discussing health/mental health history or concerns) than needing to for e.g. a non-AI search. So controlling access levels might be tricky or perhaps ultimately futile anyway.
Tasked with finding a solution to this problem, can I suggest that you ask both Ai agents if they can sort it out between them. Tongue in cheek but it might be interesting to see what happens.
I have played AI off against each other several times before. It’s like having competing analysts try and provide the more comprehensive analysis. Usually, one of them has too many errors though. And interestingly, we probably find we get to a point where people won’t know what’s wrong because they use AI too much and just assume it’s right. What’s the saying, if you know, you know. Well, if you don’t know you don’t know. Or was that Rumsfeld! haha