I dont think any of these tech execs (all execs?) use their products. They all have assistants to do everything for them, so they have no idea what this whole "internet" thing is, other than it makes them money.

The LLM? Yes, actually, and it's not secret:

https://aistudio.google.com/

The "preview" version are often pretty good, before Google deep fries them with sycophantic RHLF. For example, Gemini 2.0 and 2.5 Pro both peaked in temporary experimental versions, before getting worse (and benchmark maxxed) in subsequent updates.

But if you really want unenshittified LLMs, look into open weights models like GLM. They're useful tools, and locally runnable. They are kind of a "secret useful AI out there that plebs don’t get to use" because of the software finickiness and hardware requirements to run it locally make it difficult.

On top of that, Google employees probably have access to "teacher" models and such that the public never gets to see.

For search? IDK. I'm less familiar on what Google does internally, but honestly, from what I've read, the higher ups are drinking kool-aid and assert all is fine.

I assume all high-level positions in tech companies are using better versions than what they shove out to the rest.

I mean, Microsoft treats Enterprise users with class with Windows 10 Enterprise. That version doesn't have nearly the amount of bloat that even Professional has. Hell, Enterprise doesn't even have that stupid online search function.

So it's like they KNOW they have greenlit some shitty ideas, but, they won't deal with it so why not just throw it all onto others to make their experiences miserable?

another thing to mention, YOUTUBE. The search bar doesn't even do anything, it shows RECOMMENDATIONS instead of answers to the search.

Paying doesn't even stop that! It's actually maddening

Those assholes probably kept a working version of Inbox for themselves. 😡

Do you think the CEO of Kraft feeds macaroni cheese dinner from a box to his kids?

The problem with conspiracy theories like this is it assumed that the people doing the conspiracy are competent.

I think yes, and no. There are certainly in-house tools that the outside folks don't get. LLMs for sure have better tiers and loosened guardrails.

...buuuuut, the people at an 'executive' level also are entirely unlike you and me. They are simultaneously as gullible and foolish as the 'sheep' of society, who are also buying into the 'AI' hype of LLMs, and so far removed from our situation that even using an LLM or search engine is entirely outside of their experience. They aren't going to be using an LLM to plan out a vacation or a work schedule and have it fail any more than they would have looked through a SEO optimized bullshit website about vacuum cleaners (or super slideshow-ified list of 'top ten pacific vacations!' website to show you a bunch of ads) five years ago. They'll ask the LLM (/search engine and only look at the ai at top) for the best pacific vacations and then tell their assistant to plan a vacation for them based on a quick glance at the result (or the same for the vacuum cleaner to replace the one that broke when their house cleaner was trying to get the super long hair from the super fru-fru breed that's only allowed in two rooms in the house out of the super luxurious thick rug).

The way they use the LLM is perfectly fine for them. They aren't going to see any negatives from it, so the in-house or publicly available versions aren't really the reason for their ability to 'crow' about it. Same for the general downtrend of the internet. Their use case fucking sucks, and it isn't affected.

There is Kagi for the rest of us

No, they just have their human assisstants as a filter to use the entshittified search.

I for sure believe they do. The number of applications that launched with genuinely useful features, only to have them rolled back because of public backlash or shoved behind a paywall, has always pissed me off. Take Bing image search. When it first showed up, you could actually use it for OPSEC and for tracing the origin of suspect memes. If something felt astroturfed or a user seemed like a bot, I could verify it with Bing search. Now I can’t even use it to search a logo because they gutted it. There’s no way I believe that search capability doesn’t still exist behind closed doors, in the hands of political firms, law offices, or government agencies.

every time I see people(?) crowing about how amazing AI is

You’re correct that there’s a massive flood of bots pushing it everywhere. But regardless of what the subject is, once someone has “bought in” to a scam they tend to stick with it and defend it no matter what. Because the alternative is admitting they were fooled, and that’s basically an uncrossable bridge for most folks.

People on their literal death beds were using their literal last words (before being intubated with covid) to threaten nurses not to go near them with “the jab”. So it really doesn’t surprise me that people continue using “AI” despite it being worse than worthless for literally everything

Honestly, I wouldn't be surprised if there were some secret Google search engine that still has decent search results instead of the paid-for mess that us plebs get.

AI, on the other hand? The flaws are intrinsic.

Nah, all the SEO nobheads poisoned search well before Google managed it.

The internet has been inventing nonsense based seemingly on your search queries for a long-ass time.

No. They're drinking their own coolaid.

They've offloaded what little thinking they did to LLMs (not that LLMs can think, but in this case it makes no difference), and at this point would no longer be able to function if they had to think for themselves.

Don't think of them as human people with human needs.

They're mere parasites, all higher functions withered away through lack of use, now more than ever.

They could die and be replaced by their chatbots, and we wouldn't notice a difference.

It doesn't take a lot of tech skill to be an exec for a tech company. My guess is they fall into the same category as those that see the AI overview on Google and think "wow this is so much better" and never second guess the results.

Gemini 3 Pro is pretty nuts already.

But yes, labs have unreleased higher cost models. Like the OpenAI model that was thousands of dollars per ARC-AGI answer. Or limited release models with different post-training like the Claude for the DoD.

When you talk about a secret useful AI — what are you trying to use AI for that you are feeling modern models are deficient in?

This is really the fear we should all have. And I've wondered about this specifically in the case of Thiel, who seems quite off their rocker.

Some things we know.

Architectural, the underpinnings of LLM's existed long before the modern crops. Attention is all you need is basic reading these days; Google literally invented transformers, but failed to create the first llm. This is important.

Modern LLM's came through basically two aspects of scaling a transformer. First, massively scale the transformer. Second, massively scale the training dataset. This is what OpenAI did. What google missed was that the emergent properties of networks change with scale. But just scaling a large neural network alone isn't enough. You need enough data to allow it to converge on interesting and useful features.

On the first part, of scaling the network. This is basically what we've done so far, along with some cleverness around how training data is presented, to create improvements to existing generative models. Larger models, are basically better models. There is some nuance here but not much. There have been no new architecural improvements that have resulted in the kind of order of magnitude scaling in improvement we saw in the jump from lstm/GAN days, to transformers.

Now what we also know, is that its incredibly opaque what is actually presented to the public. Open source models, some are in the range of 100's of billions of parameters Most aren't that big. I have quen3-vl on my local machine, its 33 billion parameters. I think I've seen some 400b parameter models in the open source world, but I haven't bothered downloading them because I can't run them. We don't actually know how many billion parameters models like Opus-4.5 or whatever shit stack OpenAI is sending out these days. Its probably in the range of 200b-500b, which we can infer based on the upper limits of what can fit on the most advanced server grade hardware. Beyond that, its MoE, multiple models on multiple GPU's conferring results.

What we haven't seen is any kind of stepwise, order of magnitude improvement since the 3.5-4 jump open AI made a few years ago. Its been very.. iterative, which is to say, underwhelming, since 2023. Its very clear that an upper limit was reached and most of the improvements have been around QoL and nice engineering, but nothing has fundamentally or noticeably improved in terms of the underlying quality of these models. That is in and of itself interesting and there could be several explanations of this.

Getting very far beyond this takes us beyond the hardware limitations of even the most advanced manufacturing we currently have available to us. I think the most a blackwell card has is ~288GB of VRAM? Now it might be at this scale we just don't have hardware available to even try and look over the hedge to see what or how a larger model might perform. This is one explanation: we hit the memory limits of hardware and we might not see a major performance improvement until we get into the TB range of memory on GPU's.

Another explanation, could be that at the consumer level, they stopped throwing more compute resources at the problem. Remember the MoE thing? Well these companies, allegedly, are supposed to make money. Its possible that they just stopped throwing more resources at their product lines, and that more MoE does actually result in better performance.

In the first scenario I outlined, executives would be limited to the same useful, but kinda-crappy LLM's we all have access to. In the second scenario, executives might have access to super powered, high MoE versions.

If the second scenario is true and when highly clustered, llm's can demonstrate an additional stepwise performance improvement, then we're already fucked. But if this were the case, its not like western companies have a monopoly on GPUs or even models. And we're not seeing that kind of massive performance bump elsewhere, so its likely that MoE also has its limits and they've been reached at this point. Its also possible we've reached the limits of the training data. That even having consumed all of 400k's years of humanities output, and its still too dumb to draw a full glass of wine. I don't believe this, but it is possible.

I fully believe in old.Google.com existing now

Apple execs use iPhones with a modified iOS that lets them use a keyboard that doesn't suck and doesn't have sponsored auto-corrections, and I will die on this hill. /s

Honestly though, I think a lot of people at Microsoft just use Macs. Google? I have no idea. I would imagine the smarter ones search with DuckDuckGo or something like that. Apple? They probably just use their own stuff since they look down their noses at the competition, but they use the best. Co-founder Steve Wozniak basically confirmed that. Even though he no longer works there, he said they send him the best iPhone every year. (I'm not sure if they still do that.) He also said it's pretty but he wishes it did half the shit his Android phone does. He also uses custom firmware on Android, so he's not just using a stock Android phone, he's running something like GrapheneOS or LineageOS. So Woz is basically just collecting them at this point. Maybe he donates them. I dunno. So yeah, Tim Cook has a Mac on his desk, but it's not the $480 M4 Mac mini, it's a fully spec'd Mac Studio that would probably set you back close to 10 grand. Because why would he use anything less? And he's carrying the top spec'd MacBook Pro. And a 1TB iPhone Pro... he doesn't seem like a big guy, so he might just be using the "regular" Pro and not the Pro Max, as a personal preference. Honestly, the executives can use whatever they want, but they'll only be seen with the flagship.