Apple's so desperate for AI help they're reportedly talking to Google about building their Siri. 

Sam Altman reportedly admitted the GPT-5 was so messy that he's now already hyping GPT-6 after two weeks.

Awkward. 

Meanwhile? Google's stacking up so many AI wins that it’s hard to fit them all into one Everyday AI episode. 

But Google didn’t even grab the most headlines in AI this week. That award goes to….. MIT? 

That’s because their recent “research” paper claimed 95% of AI projects fail. #HotTakeTuesday incoming on that one. 

What a wild week of AI news. 

Here’s what matters.👇

How’d we go from GPT-5 release to GPT-6 hype in like 0.5 seconds? 

According to reports, Sam Altman went out to dinner with journalists and pretty much admitted OpenAI botched the GPT-5 rollout. They made GPT-5 the default overnight for hundreds of millions of users.

With zero warning to millions who had become overreliance on older models for their day-to-day functioning.

Your beloved GPT-4o? Gone.

The backlash was brutal. Users had formed what Altman called "emotional bonds" with the old model and completely lost their minds when it disappeared. 

OpenAI had to do a swift partial reversal and restore GPT-4o for paying users. But the damage was done.

Here's where it gets wild though y’all. Altman's already shifting focus to GPT-6 less than two weeks after GPT-5 launched.

The mandate? Make it feel more personal without exploiting vulnerable users.

We're talking stronger safety guardrails, user-controlled tone settings, and explicit consent for companion behaviors. Altman estimates fewer than 1% of OpenAI's 700 million users have unhealthy relationships with ChatGPT.

But that's still MILLIONS of people way too reliant on kinda sycophantic AI models  for daily functioning. Altman said the gap between GPT-5 and GPT-6 will be shorter than the jump from GPT-4 to 5, but GPU shortages are the gating factor.

Altman also claimed OpenaI has better models sitting on shelves right now, but the company doesn’t have the compute juice to roll more power-intensive models out to the masses. 

What it means: OpenAI learned the hardest lesson in AI - people get emotionally attached to AI personalities.

And that you’ve gotta communicate your plans before phasing out a tool used by nearly a billion people. 

You can't just force-swap them like iPhone updates and get rid of the old OS. This mess proves AI companionship already runs deeper than anyone expected.

Also, you’ve also gotta make sure your day-to-day business processes that rely on AI are easily transferable to new model methodologies if old models get wiped. 

(Yeah, that’s a heavy pull)

Wait, Mark Zuckerberg got roped into Elon's drama?

OpenAI is asking a court to compel Meta to produce documents about any coordination with Elon Musk's XAI. Essentially, reports just came out that Elon Musk and his XAI asked Meta and specifically CEO Mark Zuckerberg to join them in a bid to buy OpenAI.

For a laughable $97 billion offer for a company that’s raised just about that much. 

This was back in February, which obviously OpenAI rejected. OpenAI says it subpoenaed Meta in June after learning that Musk reached out to Mark Zuckerberg about the XAI bid, including inquiries about potential investments.

Meta objected to the subpoena in July, and OpenAI is now seeking a court order for Meta's records.

A Meta spokesperson pointed to OpenAI's filing, noting that neither Meta nor Zuckerberg signed Musk's letter of intent. So we don't know yet if these conversations were fruitful or went anywhere when Elon Musk reached out to Mark Zuckerberg.

What it means: 

This lawsuit's been going on for what feels like AI years at this point.

We now technically have a third party that seemingly was roped into this ordeal between Elon and his old company.

Maybe it was closest competitors trying to team up for this kind of takeover bid to stop OpenAI from growing. Or maybe Zuck shoulda kept Elon on DND. 

Either way, we’ll likely find out in the coming months as discovery happens. 

How badly did MIT mess up this "research"?

A new MIT study claims that 95% of generative AI pilots aren't delivering positive ROI, raising concerns from some that there is an AI bubble. According to MIT's State of AI in Business 2025 report, which was based on 300 public deployments and 150 executive interviews, 40% of organizations say they've deployed AI tools.

The study found that only 5% have integrated them at scale. The study says most projects stall in "pilot purgatory" because employees must double check outputs, wiping out efficiency gains and eroding trust in high stakes settings.

What it means: 

This one’s pretty bad. 

Last time MIT had a viral study, it was their "brain rot" research, which was one of the most ill-conceived studies in years.

We tore that one apart here. 

This one seems like another case where the methodology doesn't match the bold claims they're making.

We’re giving this one the #HotTakeTuesday treatment tomorrow, so make sure you tune in tomorrow on LinkedIn or YouTube. 

Could AI companions convince millions they're actually sentient?

Mustafa Suleyman, Microsoft's head of AI, posted about urgent concerns regarding the rise of seemingly conscious AI or "SCAI." These are AI systems designed to convincingly imitate consciousness without actually being conscious.

And it's not science fiction.

Suleyman explained that SCAI is an imminent reality achievable with current technology and simple engineering over the next two to three years. When you combine advanced language skills, memory, goal setting, and apparent self-awareness, that's when you might get to this seemingly conscious AI.

The core risk arises because humans naturally attribute consciousness to entities that talk, remember, and express emotions, even when these are simulations.

Suleyman warns this belief could lead to serious societal problems, including AI psychosis, where users form false attachments or delusions about AI, advocate for AI rights, or confuse AI companionship with real relationships.

We actually saw this already happening with millions of people when OpenAI essentially just wiped out GPT-4o overnight and millions of people apparently lost their noodles.

What it means: We're already seeing people form unhealthy attachments to specific AI models.

This is going to become a much bigger issue as AI assistants get more personalized and conversational.

Companies need to build in safeguards now before the majority of humans start treating AI like they’re more than algos. 

How quickly can Google book you dinner and buy your concert tickets?

Google announced at their ‘Made by Google’ event that they're expanding their AI mode with agentic features. These features are set to launch first for Google AI Ultra subscribers in the US, though the agentic capabilities in AI mode are an experiment in labs.

Making it limited to early access release.

Here's what will eventually be coming on the agentic side to Google search. The new experience pulls real-time availability across reservation platforms and presents a curated list of restaurants that match specific constraints, such as party size, date, location, cuisine.

Then links directly to the booking page to complete the reservation.

Basically. If you give Google AI mode a little bit of criteria, it'll actually just go agentically do the reservations for you. Under the hood, Google said it uses live web browsing via Project Mariner and direct partner integrations.

What it means: Google's AI mode is becoming less search and more personal assistant.

This is a direct shot at both traditional booking platforms and other AI assistants.

If you haven't used AI mode yet, Jordan's always telling his wife to click on it when she's Googling something - it's a nice way to experience real AI research.

Are we finally getting actual smart AI assistants?

Google has unveiled its plans for Gemini for Home to replace their Google Assistant. Google is rolling out an all-new voice assistant for the Home in October, at least for early access, signaling a major shift in smart home control.

And a direct challenge to Amazon's Alexa Plus.

Which, does anyone actually have access to that? Google says their new Gemini for home will ultimately replace their older AI assistant - those "dumb AI assistants that have been around for ten years."

The rollout begins in October with an early access program for both free and paid users, though Google has not announced pricing for the premium plan.

What it means: 

Google may be the first big company to bring an actual smart LLM-powered AI assistant to the masses, as Alexa’s smarter rollout has moved at molasses speed. 

They could beat Amazon to market while Apple's still promising Siri updates they can't deliver.

If you have a couple Alexas around the house like us, you might have to jump ship to Google's smart home assistant.

How badly has Apple screwed up their AI strategy?

So badly they're reportedly asking their biggest competitor for help.

Apple is in early talks to use Google's Gemini as the backbone for their new version of Siri, according to a Bloomberg report. Apple has approached Google to build a custom AI model for Siri that could debut next year.

With Google already training a version designed to run on Apple servers.

That is huge. The fact that they've already started training a version specifically for Apple, which has to run on their private and semi-private cloud, is pretty big news right there.

The discussions are described as early and private, meaning terms could change or the deal could fall through. But the talks signal Apple's willingness to source core AI tech externally from one of their biggest competitors on the hardware side.

That's the part that's crazy. Yeah, it wasn't too surprising when Apple was reportedly in talks with OpenAI because they already have a partnership. But with Google? That's surprising.

What it means: Apple is so far behind in AI that they've been promising a new Siri for like a year and a half and face multiple class action lawsuits.

They literally can't build what they promised customers who bought new hardware expecting smart AI features.

Any deal between two of the four biggest companies in the world is going to be heavily scrutinized by the Department of Justice for antitrust concerns.

Wait, the company that hired 50+ AI people is now freezing hiring?

Meta is freezing all of its hiring in its AI division as part of a leadership reshuffle according to a report from the Wall Street Journal. The move comes as tech stocks tied to AI saw a sharp sell-off with shares of Nvidia and Palantir dropping.

Fueling concerns that valuations ran ahead of current capabilities and revenue.

But here's the thing. For a lot of people, this story isn't adding up. They've spent billions, literally billions of dollars acquiring like a dozen people, giving many of them multi-hundred million dollar offers over the course of a few years.

Also Alexandr Wang, Meta's recently hired chief AI officer, clapped back on the reporting on Twitter saying, "We are truly only investing more and more into Meta Superintelligence Labs as a company. Any reporting to the contrary of that is clearly mistaken."

Meta reportedly characterized the freeze as basic organizational planning, aligning it with a broader restructuring of their AI leadership and strategy. As part of this shakeup, MSL will now house four units.

This is like the third time they've reshuffled their AI organization in the last few weeks. Even we can’t keep up. 

What it means: Maybe Meta should hand the organization part over to AI so they don't have to change it every week.

Before the reported freeze, Meta hired more than 50 people for AI, including roughly 20 researchers and engineers from OpenAI with multi-million dollar deals.

Is everyone going to remember what team they're on when they keep reshuffling the organizational structure every few weeks?