> We used novel synthetic data generation techniques, such as distilling outputs from OpenAI o1-preview, to post-train the model for its core behaviors. This approach allowed us to rapidly address writing quality and new user interactions, all without relying on human-generated data.
So they took a bunch of human-generated data and put it into o1, then used the output of o1 to train canvas? How can they claim that this is a completely synthetic dataset? Humans were still involved in providing data.
It may be related to the famous 301 view count where YouTube would stop updating the total pending further verification that the views are legitimate. See: https://youtu.be/oIkhgagvrjI This behavior was later removed in 2015, but I wouldn’t be surprised if something similar is happening here.
This bot answers any prompt, even those completely unrelated to the product (ex: generating code, writing paragraphs, etc) I imagine this could be abused and accumulate unwanted costs. Also, how do you ensure the bot isn’t subject to hallucinations? I would never risk giving a customer wrong information about my product.
How many more platforms do we need where users can post all sorts of media and follow other’s content? Every social media site copies features from another, and they are all basically the same item wrapped in a different package. When will people wake up and return to RSS?
Feels like they were too eager to scoop up Twitter refugees. The platform is still missing basic features and new users turn away immediately. Nobody’s going to join a platform and then keep their account open for 6 months while the platform plays catch-up. The launch would have been much more successful if they waited another month. I understand this is a peak time to capture former Twitter users, but in the long run it may have been a mistake. Will be interesting to see how they market it as things develop.
This is stupid. Who on earth would want to pay a subscription for a PC when they still need to buy one to stream the OS?
I see the value proposition, but it’s only as fast (and expensive) as your internet speed.
- Businesses and consumers could buy less expensive hardware, and simply pay $x/month for access.
- IT could remotely troubleshoot
- Fewer hardware issues
- A $10/month subscription would last longer than a $1200 laptop for most consumers.
- MS handles all software updates, stores all your personal data, has complete control of hardware and software.
- Some sort of device is still required, how does it respond to input from a keyboard, mouse, controller, etc? How do USBs, Raspberry Pi’s, External storage devices work?
- You still need a screen. I could see this being an app on smart TVs and mobile devices. (The form-factor of a powerful PC is smaller because of the streaming aspect)
- Internet is still relatively slow, every input would feel laggy and unresponsive.
- Anyone can guess your username/pwd and gain access to your PC.
- Hackers now have a single target.
- If Microsoft’s servers ever go down, you can’t use your PC at all.
- You are subject to any future government regulations with no choice to opt out.
- You lose access to everything if your subscription expires.
- MS has an unprecedented level of access to data that will inevitably be used for advertising.
- MS becomes your ISP and can filter traffic however they want.
I’ve been using 12ft since beta. They’ve been hit by some legal notices and no longer work on some sites like NYT. It’s just a game of whack-a-mole. Ideally these workarounds are open source so they can’t just shut a domain down.
Not OP, but I loved Neeva’s results. That being said, it’s a bit too personalized to my liking. For example, it automatically detected my location to display the weather, and the app’s search history cannot be disabled without private mode. For a “private” search engine, it did way too many personalizations by default. These few things, and a somewhat confusing UI made me switch to Orion (Kagi).
> And employees are opted in without any notice from their employer
[edit] Obviously professional communication channels are monitored, but it would still be nice to be informed when policies change. (Anyhow, personal and professional communications should always be kept separate to avoid this issue)
Been using on iOS for a few months now. I enjoy using Orion because it actually gives the user options and imo fixes the issues I had with the Safari redesign. Love that it’s fast and intuitive.
One thing I don’t like is that there’s no option to never save history (like the DDG app)
Every time I try to leave HN for some other front-end, there are subtle design features I miss. Every button on HN has a purpose, is easily accessible, and works as expected.
There’s a quote that good design is not when there’s nothing left to add, but when there’s nothing to take away. HN follows this perfectly.
Sure, I could have dark mode, previews, and notifications, but I’d lose elsewhere unless I built my own front-end.
Because publishers don’t know the difference between news and noise. Most long-form articles can be summarized in 1-5 bullet points when all the fluff is removed.
I used to have an app that curated topics into short summaries. It was curated by humans but they failed to monetize and shut down.
As someone running a side project, I definitely agree with this. I built up an array of skills that don’t have much depth, but it allows me to follow along with different topics when they come up.
The “M” strategy will give you a nice foundation in a few core skills, while being able to contribute elsewhere.
So they took a bunch of human-generated data and put it into o1, then used the output of o1 to train canvas? How can they claim that this is a completely synthetic dataset? Humans were still involved in providing data.