Checking the bus factor:
Btrfs regulars:
- 1 from Meta
- 1 from Oracle
- 4 from SuSe
- 2 from WDC
bcachefs:
- Kent Overstreet
1,696 karma · joined May 6, 2022
Btrfs regulars:
- 1 from Meta
- 1 from Oracle
- 4 from SuSe
- 2 from WDC
bcachefs:
- Kent Overstreet
Not if you use git-annex.
GPT-3 is probably one of the last models trained without synthetic data. Meanwhile gpt-oss is rumored to be only trained on synthetic data.
I also miss the 2020 GPT-3 based AI Dungeon fine tune.
The AI labs and the downstream companies that sell training data to them vacuum up everything they can.
Illegal residential proxies (botnets) that once have been used by hackers and scammers are now used to vacuum up the Internet.
They are now vacuuming up antique books that are practically useless.[1]
In face of this is is unthinkable to me that they are not training on API data.
> The business loss of trust would outweigh any benefits of the data.
The loss of trust is already here.
I know of one German company that uses AI only in areas where they have to compete with (foreign) startups. For their core business and everything else they are waiting for an on-prem solution. Apparently Microsoft can provide on-prem GPT-5.
[1] https://lesekauz.de/forum/thread/1999-sammelbestellungen-von...
With Chinese providers at least I'm getting a open weight model out of it.
> WARNING: workerd is not a hardened sandbox
> workerd tries to isolate each Worker so that it can only access the resources it is configured to access. However, workerd on its own does not contain suitable defense-in-depth against the possibility of implementation bugs. When using workerd to run possibly-malicious code, you must run it inside an appropriate secure sandbox, such as a virtual machine. The Cloudflare Workers hosting service in particular uses many additional layers of defense-in-depth.
Sandstorm was great because it did proper sandboxing. This is pretty weak by comparison.
Google Gemma disproves this
Can you explain this more precisely?
The Trump admin is now regulating frontier models though.
> [...]
> Efforts to block the Knickebein headache were codenamed "Aspirin".
Love it.
This gave me a chuckle, "PhD-level code" is gross actually. Have you ever looked at the code of research papers?
My favorites:
> 6. Structured on-device integration
> AI services will be able to easily interact with other apps installed on the device and perform tasks on behalf of the user within those apps, for tasks that the apps and the user have chosen to make available to AI services. These tasks include “send a message”, “create a note”, “schedule a meeting”. This includes access to certain Google apps (i.e.Gmail, Calendar, Drive, Docs, Maps, YouTube, Messages and Phone) that Alphabet will make available through operating system-level integration channels.
> [...]
> For instance, Android implements structured on-device integration through App Functions, which developers can enable for their apps, and which can be accessed by AI services without being reserved anymore for Google services, such as Google Assistant or Gemini.
> 7. Screen automation
> AI services will be able to automate multi-step tasks within apps, on behalf of the user upon their consent. They will do so by imitating user behaviour in a separate virtual window, which makes it possible for the assistant to complete the task in the background, while the user can do something else. [...]
> Android implements screen automation via Computer Control, which can automatically access apps, and which is currently reserved for Google’s services, such as Gemini.
> 9. System-level on-device models
> AI services will be able to call on existing on-device models (“ODMs”), including the Gemini Nano ODMs, that are part of the DMA designated operating system, already preinstalled on Android devices and already made accessible to third parties. As a result of the measures, third-party AI services will have guarantees of equal access (for example, in terms of performance) to ODMs, as Google’s services. [...]
> 10. On-device model implementation
> Third parties will be able to install, run and use on-device models (ODMs) under the same hardware‑resource and background‑execution conditions that Google’s own models enjoy, and will allow their ODMs to be shared centrally with other apps. [...]