> Upon analysis of the new data, we discovered that it included unattributable Notion AI chat transcripts used for the Insights feature. Chats weren't labeled with who wrote them or which workspace they came from, and the transcripts did not contain any tool call results. Beyond the words in the chat transcripts themselves, the data contained no information that could specifically identify the workspace, user, or company.
> As a precaution, we recommend that you treat this unattributable data as if it were exposed. If anyone on your team pasted sensitive information like passwords, API keys, or other credentials into Notion AI during this timeframe, we recommend rotating them right away.
Okay, so the actual impact is quite interesting: they did not notify you if they could not attribute the transcripts to you! You should really treat anything sensitive sent to Notion AI as compromised because they are not sure where these transcripts came from.
> treating API tokens as compromised the moment you pasted them into an AI chat seems a reasonable default in general.
Coding agents in my experience alert you to this, which is good. I pasted what was temporary token I made into Claude, just to test something, and after it had done the task it said something like, "you gave me a live token to work with in the chat, I'd advise you to treat that as exposed and rotate it."
It would be a good standard for all chat interfaces to have those "alerts" built in.
I had a similar experience yesterday with Argon (in Antigravity), but unfortunately it didn't come up until I explicitly asked it for a security audit. It had previously stored a secret API key in plain text [because it thought it made local testing easier], and didn't warn me before I'd committed to GitHub and deployed to Vercel. So, 50/50 but I've still added a note in a skill file to never do this again.
That’s like leaving your phone in the public square. You should expect that it could get stolen. But Mike’s friend can’t go “well to be fair...” when Mike steals it.
(It's fine, of course, to use AI tools—we use them heavily ourselves. Just please don't use them to generate or edit text to be read by fellow humans on Hacker News.)
It's impossible to know for sure what was LLM generated or edited, but we have to rely on imperfect classifiers or else the site would be completely flooded by genai text, which is not what the community here wants to read, nor in the intended spirit of HN.
The author and submitter of this post also previously submitted an app that makes the claim that if AI wrote it, you shouldn’t have to read it: https://news.ycombinator.com/item?id=49351138
I really want an evaluation of all of these classes of remote desktop systems and latency. Almost all of my remote display experiences have been mediocre due to lag. Maybe my tolerance is too low? How is the latency on this product?
Only RDP has felt truly responsive to me, and my understanding is that it cheats by not sending raster graphics and instead drawing primaries directly on the remote system and then updating the state based on the server?
Yes, RDP is actually more like sending commands to a remote GPU. VNC and the others are, as far as I know, just doing screen-capture and then possibly some compression on the resulting video stream.
The RDP specifications are partially documented by MS, and even the publicly available parts are huge.
Sunshine/moonlight combo is still best streaming I've ever used, both in term of quality and latency. Another best thing is the client is available on many platforms.
I tried Sunshine and Moonlight and with a macOS host, colors had to be manually fixed, and then subsequently macOS hides the cursor when typing and bringing it back caused real issues. I think this is a known defect. I used the local cursor option and got pretty close but the initial setup was rough. It is clearly intended for gaming in behavior.
The main issue I think is that my poor M2 Air cannot run iOS simulators and Sunshine simultaneously without thermal throttling because of the video streaming aspect. It did feel somewhat reasonably responsive on a LAN, but I can’t say over more unreliable networks.
I agree, I have benchmarks planned for the future.
As far as the latency on MeowDisplay, In my testing on Wifi, I measured around 14-17ms median from capture to receiver handoff, not e2e. I've also seen occasional spikes around 50-90ms and some stalled frames, you do notice the stalls but overall it's pretty responsive.
I've preferred to show the cursor as it is on the mac instead of rendering it locally on the phone which may make it feel smoother but the trade-off is that it may not accurately represent where the cursor actually is on the mac if things lag or hang.
That's pretty much been the only compromise I've had to make when migrating to an entirely Linux ecosystem at home. I settled on NoMachine as my pick of 'remote control desktop' software after trying a whole lot of other options (including various flavours of VNC).
Every time I use it I'm reminded of how good RDP is/was.
No, I'm not sure if I've heard of that one. Will check it out.
Immediate reservation about Parsec is that it requires internet access to its cloud services for authentication, session discovery, and connection setup. For me, that immediately sucks, but I'm still going to give it a go to see how low latency it is.
I have used Sunshine/Moonlight for gaming in the past and it was fantastic, but for whatever reason it was janky to use for remote desktop.
Assuming you are remoting into machines on your local network, I have made the Very Dumbest Compromise that I actually love, which is just leaving them connected to displays natively and creating a fake display on the controlling machine that runs a fullscreened remote desktop session set to potato. It's the only way I'll even bother to use a mac anymore, I'm not wasting another second fighting their godawful mouse input.
I really wonder why nobody even bothers to write their own marketing copy for even hobby projects like this. Is there no voice behind the person who decided to build this? What inspired them? Why? It’s just the usual AI language Claude wrote.
Well, I just probably am not going to use a software product named “bong” but I suppose that’s just me.
I think the issue with running something like this or SearXNG is that the good indexes will fight your efforts to use them unless you pay up and use the API. This product also suggests the google smart suggest API in the default config, which isn’t very much of a Google decoupling.
I run SearXNG (and Crawl4AI) for my self hosted librechat (self hosted chatgpt equivalent that can search, use Internet etc).
Duckduckgo.com and bing tend to respond well as well as Staan after an account is created with them.
But what we need is competition between few large search providers, because prices now are bonkers. The only large-ish search provider that doesn't just repackaged Google or Bing is Brave, but they charge $5 for 1k requests. That is just silly. 1k will not last more than two dozen uses (each chat fires off 10-20 searches).
I really think the main factor in determining if prediction markets are valuable is their predictions from outside election day. Once the “blue tsunami” was “expected” most of the markets shot overwhelmingly in one direction as the night wrapped up. You need to have a boundary for comparison that isn’t influenced by election night. Starting at 5AM is a decent cut point but I would argue the final poll date is the best comparison.
Love or hate PolyMarket, it should be on this list too.
iPhone can’t boot Linux. But there is the Macbook Neo, which runs the A18. You would need a low level exploit to compromise the chain of trust very early and the Asahi bring up work doesn’t include the touch and mobile phone components needed for iOS.
I'm genuinely curious as to how fast this was handled by Apple devices, Linux, etc. With enough advance notice, I don't see this being that terrible to patch. Was it lack of advance notice?
BC: March 2
Alberta: April 20
Manitoba: Sep 17
Western Ontario: Oct 13
Manitoba didn't give much advance warning. Western Ontario is on the same time zone as Manitoba so it was very confusing there until at the last second it was clarified. So Microsoft has an excuse for screwing up Kenora. Vancouver though should not have been a problem.
I'm in AB and work at a Windows/MS-based shop, and haven't run into any real problems with any Windows/Apple/Linux systems with the time change.
Our IT department ran a couple scripts last week and everything this week has been smooth sailing.
Things I have run into problems with have been my smart thermostat (ecobee, now fixed), my Philips Hue lights (most of my automations are via Home Assistant, but I have a handful that rely on colour temp in the Hue app), and my Ford car. (Auto time zone and auto time aren't separately configurable, and auto time zone puts me in the wrong time zone.)
No issue at all for Manitoba time (or Saskatchewan time) on Linux devices. Teams in a browser (Linux) failed though, badly - so this is a systemic Microsoft problem.
I am also curious to know if having the system clock in UTC and storing times using UTC, which I understand UNIX-based operating systems do, avoids these types of issues. It should make them easier to manage, is my guess.
A) you need the new timezone definition available and pushed everywhere regardless. It doesn't matter what zone the RTC uses.
B) For future events, you need a way to manage what happens when the time zone they're scheduled in changes OR when they change to a new time zone. Microsoft has been terrible at this part on every time zone change I've been aware of (it's hard, but it's not this hard, and the Canadian change was published with sufficient notice, and time zone changes happen for recurring events, this is not a new requirement). But there's no connection to the time zone of the RTC here either.
If you have issues with poor behavior specifically around transitions, that could be related... Repeatedly moving forward or backward for DST is something where keeping the RTC in UTC would help.
For past events, you can certainly store them as UTC and (possibly implied) location, provided that you also have a historically accurate database of all past time zone changes, so that you do not report that a meeting took place at 9:00 when it actually happened at 10:00, but the current time zone rules (at that location) are not the same today as they were on that date in the past.
For future events, you have to store them as location + timezone-naive dates and times, a.k.a. local time. Since if a meeting at a certain location is scheduled at 9:00, which is currently at 02:00 UTC, and then a timezone change happens, and UTC 2 suddenly means 10:00 at that same future date and location, the meeting will still happen at 9:00 local time. You cannot store future events as UTC only.
UTC brings its own batch of problems when it comes to local time relative scenarios.
Last time I looked best practice for a real calendar app was to store in local time with time zone, not UTC with offset.
UTC is time in theory. Points in time separated by numbers of seconds. Great until the reality of time being a social construct breaks your assumptions. Still, use UTC until you are actually building a calendar app. Depending on your use case, consider storing timezone instead of offset.
Either UTC or TAI should be used for anything that is truly a "time".
The "time" at which future events are scheduled by humans (i.e. not the time when future events are predicted, like astronomical events), is not a "time", but it is a different data type, which must have a different name and which must store different information (i.e. legal date and time at a certain location).
As another poster already said, UTC is the correct choice for any kind of timestamps and for the storage of any time information about the past. (Actually TAI is an even better choice for these purposes.)
For future events, the "time" when they will occur is not "time", but a different data type, which should be called "legal time", "local time" or the like.
The "local time" for future events must include the geographic location attached to the event, from which the applicable time zone can be derived, at a time close to when the event is due.
my experience on apple devices in alberta hasn't been any better, microsoft is just the bigger target for articles like this because more people use it for scheduling meetings.
it took around 40hrs for my apple watch and my iPhone to start agreeing on what time it was.
> I equate it to partnering with AT&T on the first iPhone.
In terms of strategy, they picked the underdog carrier (Cingular) specifically to avoid having to bundle bloatware and carrier customization at the cost of exclusivity. Likewise, when Apple brought the iPhone to Japan, they did the same thing with the underdog carrier here, SoftBank.
The strategic importance of this was as a bargaining chip with other carriers and fundamentally gave them a lot of control they would not have had if they had gone with Verizon or similar.
> As a precaution, we recommend that you treat this unattributable data as if it were exposed. If anyone on your team pasted sensitive information like passwords, API keys, or other credentials into Notion AI during this timeframe, we recommend rotating them right away.
Okay, so the actual impact is quite interesting: they did not notify you if they could not attribute the transcripts to you! You should really treat anything sensitive sent to Notion AI as compromised because they are not sure where these transcripts came from.
reply