Hacker Newsnew | past | comments | ask | show | jobs | submit | HenryMulligan's commentslogin

Apps Only mode still has plenty of ads on the home screen though.

Sounds like OpenAI might be trying their hand at TPUs, like what Google has. They are one of Google's biggest advantages in AI right now. It would also give them insurance against Nvidia being everybody's hardware supplier.


This actually points the opposite direction, to doubling down on commercial GPUs.

NVIDIA recently told their board partners that they will need to source their own RAM and will not be bundling it with chips anymore.

If there is a supply crunch on DRAM, commercial GPU production lines will start having idle downtime. That is literally the worst possible thing that can happen to a company that has invested heavily in tooling and they will negotiate at or below cost production runs to fill the gaps if a customer can bring their own DRAM to the table.


What do you mean “might be” trying their hand? It’s been widely reported that they are doing exactly that. It’s even known that they are doing it with Broadcom.

https://www.theregister.com/2025/10/13/openai_broadcom_deal/


Both of those write a single ISO to your USB stick, while Ventoy allows you to store numerous ISOs in a folder on the stick and choose which to use at runtime. Also, you can store other files like normal with the remaining space on your stick.


Data privacy and security don't matter? My secondhand RTX 3060 would buy a lot of cloud credits, but I don't want tons of highly personal data sent to the cloud. I can't imagine how it would be for healthcare and finance, at least if they properly shepherded their data.


For most people, no, privacy does not matter in this sense, and "security" would only be a relevant term if there was a pre-existing adversarial situation


Ignoring what this model architecture could do and just considering what this model does do, why would I (or anyone) want to run this model (locally) to do <insert use-case>? Is it entirely a proof-of-concept for future training on medical data? Are they looking to use this to attempt to ethically justify training on (free-tier) user's personal data via the application of noise to the training data?


You can hide that you pirated content for training


You can't hide that. You can't use technical measures to hide from discovery.

I think an entire book is a little too large to mask with this method and still end up learning anything.


U can avoid book publisher lawsuit which Anthropic is dealing with using this approach


It's the last option.

The whole framing of DP is:

Probability that you reveal private info is same whether or not you train on a particular users data.

It is useful in many cases, but google the product company specifically is going to use it for ads.


The purpose is research


Is it too big a leap for me to assume someone is going around using password spraying or whatever to compromise neglected accounts for use as spam bots?


I haven't gotten around to trying it, but this may be a solution for Reolink cameras (that don't already support RTSP/ONVIF): https://github.com/QuantumEntangledAndy/neolink


The factual info in your reference "[0]" makes no sense. The writer (and illustrator) of the Dilbert cartoon is Scott Adams [1]. I have no idea what the name you referenced has anything to do with it, other than some Japanese software. Or was all of part of your comment written by an LLM?

[1] https://en.wikipedia.org/wiki/Dilbert


Saying Hatsune Miku made X is a common joke on Twitter/Tumblr when the original person to do so has reprehensible political beliefs.


Huh, TIL. That would have caught me, too.


That whooshing sound you hear is the joke going over your head.

As for my comment being LLM slop, it's not. LLMs are way too corporatized - yes, even the local ones - to put in-jokes or references in citations.


I remember receiving this video ad for a Chevy pickup truck in a print issue of Popular Mechanics circa 2015 [0]. When disassembled, it consisted of a screen, battery, speaker, and circuit board. What the article doesn't say is that the circuit board had a micro-USB port which when connected to a computer mounted the internal storage as a drive, with the four-or-so videos it played accessible. I actually managed to find an existing home video video with the correct format (I wasn't familiar with FFmpeg at the time) and when one of the internal videos was replaced with mine (with the same filename), it would play instead. I believe the micro-USB also charged the internal LiPo battery as well, as I don't recall being worried too much about battery life. I probably still have the thing somewhere! [0] https://www.tubefilter.com/2015/04/16/chevrolet-video-ads-in...


That's closer to what I remember, but I don't think the advertisement was about a Chevy Pickup.

Apparently the group that did this advertisement was: https://www.americhip.com/ourwork/?product=product-magazine-...

And they have a whole slew of magazine inserts that they've done. I'm convinced now that I saw one of these magazine inserts, but not the one on the Chevy. Thanks for helping me track it down! I'm really close...

EDIT: I found it!! https://www.americhip.com/ourwork/microsoft-wi-fi-hotspot.ht...

It was this advertisement, I remember the cover. Sorry I wasn't very specific earlier, but I didn't realize how many "throwaway computers inside of magazines" that there were. I guess I assumed this was a one-off event with Microsoft's 365 WiFi cloud (now that I remember the chip, lol).


It is interesting that T-0 says "Excitement guaranteed" instead of "launch" or something. Besides being funny, it seems like they are officially accepting the reality that it may not go as planned (even at so early a stage), but will be thrilling nevertheless!


They said that for the first one too.


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: