Hacker News Reader

Comment by kjellsbells | Hacker News Reader

Heres something I just don't get. Microsoft just got their ass handed to them by the US Government because os (lapses in) security. Extremely more than coincidentally, Satya Nadella told the entire Microsoft org that if anyone had to choose between features and security, to choose security. I'm hearing from Microsoft people that all product roadmaps are deferred for a few months while security features are addressed. Their whole corporate spiel is "Microsoft runs on trust" (see the famous standards of business training on youtube).

And then someone goes and invents Recall. This is not the work of a lone engineer and a principal PM fishing for Impact or whatever they call success at Microsoft. This had to have gone through multiple levels of review. Microsoft PMs, CVPs, their corpo legal people, marketing approval. And yet no one stopped to say, "wait, this could blow up in our faces"?

View on HN

I'll take a shot (not at MS, so I have no inside info).

GenAI is hotter than anything else right now. As Satya publicly stated, "we made them dance" -- which shows how high-priority it is to maximize "AI innovation" at MS.

If you disagree, think about how badly "Tay" blew up in MS's face and yet they still went ahead and bundled OpenAI LLM tech into all of Office365, just so they could have bragging rights about beating Google to it.

At this point, it's a race (to where? who knows) and no Big Tech Corp wants to be seen as "not at the forefront".

That's my $0.02, anyway :)

You are absolutely right. This AI race has been sort of funny to watch among the big tech incumbents.

Google kept on launching faked demos and hurriedly released an openly race biased image generator all in a bid to catch up with OpenAI.

Meanwhile Apple has been sort of lethargic. Recently annouced a deal with OpenAI to add GPT to their devices. They seem happy to continue playing catch up in this regard.

I think Meta is probably the only big tech giant that has kind of got their execution right straight from the jump. Can't point to any slip ups from their "AI announcements".

Note that Meta aren't trying to sell AI services to most people (with the exception of advertisers, for whom it kinda makes sense).

Much easier to avoid messing up, given this.

Llama 3 is available and in your face when you open WhatsApp these days. They're very much trying to get everyone to tell them / the AI their secrets.

You forgot Amazon eating paste in the corner.

> They seem happy to continue playing catch up in this regard.

Catch up with what? Does anyone else have an AI product actually worth having?

Is it somehow bad that Apple hasn't also released an AI that doesn't do what it's supposed do half of the time?

Apple plays “catch up” by quietly hunkering down for a few years and then suddenly leapfrogging everyone else.

Except they totally flopped with virtual reality.

Mixed Reality... the vision pro is not suitable for 80% of the VR use cases ie. Gaming/porn.

> Meanwhile Apple has been sort of lethargic.

If by lethargic you mean OCR of every single image on my computer so I can copy and paste text from anything, all running locally on the NPU on my M1 chip, then lethargic is A-OK.

That's what Apple does: they ship useful features. We'll see tomorrow if they fall in the trap and run in this race to nowhere.

I tried to use AI on (intentionally?) badly formatted bank statements for a family member. Tried Apple, Google, various startups. Nothing could do it. The easy stuff is easy (raw OCR) but the hard stuff (layout recognition and reformatting) is not yet doable.

>OCR of every single image on my computer so I can copy and paste text from anything, all running locally on the NPU on my M1 chip

This functionality is baked into Windows since 10. But for some reason you need a third party program to expose it.

https://learn.microsoft.com/en-us/uwp/api/windows.media.ocr?...

https://github.com/TheJoeFin/Text-Grab

Check out Microsoft PowerToys' TextExtractor.

Apple, as it exists now, is not much an innovator. They don't need to be. They can release improved versions of the products everyone already has. Best not to mess with that by introducing half-assed tech that breaks the user experience. Wait till its matured a bit then release a smooth and polished apple version.

I think this is for the better. I’ve been using LLMs pretty regularly lately for some side projects, and they’re interesting but they’re not particularly nice products. I’m not aware of anyone who has a legitimately polished offering. There’s definitely a lot to the tech, but the signal to noise on essentially every model I’ve tried is far below normal standards for feature reliability and quality.

> hurriedly released an openly race biased image generator all in a bid to catch up with OpenAI.

If you think the bias was due to timing constraints and not intentional then you are extremely naive.

What's so hard to get? Security and trust are lip service. Corporations lie to make a buck. How is it not obvious?

Actions speak louder than words.

The user data is just too juicy a potential profit center to pass up on. So many models to create, to theoretically replace many of us.

We should assume that everyone is on their worst behavior going forward. Zoom, Slack, MS, Apple, etc.

Untrue. Employees are asked to focus on sec. The org tells you that the company runs on trust. This is not lip service - it's not that internally they state otherwise, and it's certainly naive to think so.

What happens is that, meanwhile, business is happening. People are making (small) decisions that adds up to big things, and leadership tends to trust those below to make the right calls and implement technology correctly. Which, time and time again, doesn't happen.

Yes, maybe it need to be more of a priority, but, it also isn't a conscious decision.

the reason they did this is because if they succeed in sneaking this out, MS would have been 'drinking the milkshake' of google, adobe, facebook and everyone else trying to do the same thing at an app level. it would be the holy grail of training data -- they could replace or simulate the entire user at an inter-app level for A/B testing anything they want, or worse. the idea of having full user behavior history of the majority of the worlds computer users is just too tempting not to try for them and im sure theyll try again despite any backlash.

People have been “training” models on windows api calls for decades.

> Heres something I just don't get.

> And then someone goes and invents Recall

Maybe you should read about the history of Microsoft, especially about its security.

But people forget easily.

Wasn't recall out in preview before the whole security snafus happened?

No. Microsoft very frequently puts out new features into the normal releases of Windows 11 at the same time as insider testers get rollouts.

Wasn't recall out in preview before the whole security snafus happened?

Reply on news.ycombinator.com