1) The designation has nothing to do with hacking, either from Anthropic or otherwise. The dispute is about Anthropic barring the DoD from using Claude to develop autonomous weapons and conduct domestic surveillance.
2) The DoD does not care about any of the hacking OpenAI does as long as the DoD can do what they want with ChatGPT.
3) Historically, the US government has been interested in hacking other nation states, friendly and not. So OpenAI being able to hack Australian systems is a selling point.
I mean, everything should eh sandboxed imo. Even OS included software.
Problem with sandboxes is that it doesn't solve the "user who doesn't know or care" problem.
If you have to give access to particular files to an app, your average user will just give access to all. Someone installing tiktok doesn't look at the permissions...
The best thing would be for apps to operate on their own copy of a file regardless. But that doesn't stop bad apps from leaking importsnt files. Unless files can be signed as being for x thing requiring y special permissions as granted by the originator and not the user.
The crowd on here we can't to be able to truly own our devices. But I think for the safety of the average person they should be locked down by default, same as dev mode on android devices where you accept liability for doing silly things.
OpenAI's business model would align infra as a cost center rather than infra as a profit center (e.g. Google / AWS). Perhaps there's something there. I'd say also the OpenAI as a grad school that just happens to have a business aspect is also part of this. Bringing a tonne of good process on top of the build fast break things startup stuff would have cramped research speed significantly.
It's likely that OpenAI has gotten as good as it is because it ignored the traditional sysadmin stuff and went scrappy.
I worked there, but this is just my opinion and guesses, not facts.
For me the most useful thing about plan mode in Claude or Codex is not the plan, I usually barely read it, it's that it forces models to gather as much context as possible before implementing. Without plan mode models tend to implement wrong solutions because they are reading minimal context before writing code.
Of course this can be done with just another prompt, but that's basically what plan mode is. It's nothing more than a pre-defined prompt in the harnesses.
Sure, dude. "Women get bitten by the vagueness of their own rules, since they want to keep all options open" sounds like a completely detached, "I have no skin in this game" argument. Eyeroll.
> This library solves that problem. It registers xarray data structures as custom JAX PyTrees. This allows JAX to seamlessly flatten xarray objects into their raw arrays for accelerated computation and then unflatten the results back into fully labeled xarray objects, preserving critical metadata like dimension names and coordinates.
flatten and unflatten with datatypes is necessary for unrolling loops for performance.
Which is the correct logic for probabilistic logic, for expressions with frequentist or symbolic distributions as values? Are quantum logic and quantum statistical mechanics the appropriate or useful tools for all probabilistic logic?
dist_a1 <operator> dist_a2
uncertainties does mean±dev in Python with numpy types.
> W3C CSVW supports per-column schema. ( with URIs for datatypes )
> Serialize a dict containing a value with uncertainties and/or Pint (or astropy.units) and complex values to JSON, then read it from JSON back to the same types. Handle datetimes, complex values, and categoricals
IEEE-754 specifies NaN (null), ±0, three infinities (positive, negative, and unsigned), but IEEE-754 does not specify a representation for categoricals, datetimes (like ISO8601), or complex numbers.
XSD (XML Schema Datatypes), which RDFS vocabularies often use to specify the rdfs:range of an rdfs:Property, does not specify how to specify abstract complex numbers; but OpenMath RDF, and QUDT (Quantities, Units, Dimensions, and Types) and OM Ontology all have a way to save complex numbers to disk, too.
From the same article : “what else would you call the account of a woman who was arrested by incompetent police based on nothing more than a single Flock camera piece of data, and held in jail for 13 days”
Could you elaborate on what you mean by "don't interact with hardware much anymore"? If you compare modern drivers with similar ones from 20 years ago, for example, you'll find the modern ones tend to be quite a bit more complex.
It's funny how history repeats itself... from 61 years ago:
Pick OS had multi-value fields... Not quite the same, but it was the "Excel", and so much more, at the time.
Absolutely not all cops and prosecutors, or even the majority of them, are this lazy. Every tool is a tool that lets them "be event lazier" and make it "easier for them to abuse us". Removing any tool that let bad-actors be lazy is not an alternative for removing the bad-actor themselves.
This situation the equivalent of an eye-witness saying "Ya, I think it might have been this person I know, Tom!" and a cop/prosecutor taking that at face value and arresting Tom (and the system that allowed that to happen).
The people who make decisions and take action so (grossly) negligently need to be held accountable and face consequences. Blaming the tool (whatever tool it is) is giving the people who have the actual responsibility an escape hatch to avoid accountability.
Depending on where you draw the line for millennials the school computers could have been running anything from Windows 95 to Windows XP. BASIC wouldn't have been included in XP or 2000, and depending on which BASIC they were using, it may have been easier to use something with more features like syntax highlighting. Languages like python, perl, or java can be used to teach the same concepts and would have been more applicable to what people were actually coding with at the time.
The statnews article only mentions that the coroner said the death was "with" measles. The Ars Technica article includes the update that the coroner added, that the death was "from" measles. It also goes into detail about why a ruptured spleen implicates measles.
Just wanted to post this here, someday it might be cool to have this regionally as well, but Sep 12-14 HomelabFest will be in St. Louis, MO (technically St. Charles, which is closer to the airport). It will include a number of homelab-related content creators, talks, a bunch of cool exhibitors (mix of community and companies in the homelab space)... hopefully a good time for all!
I meant to post this earlier but was a bit overwhelmed with all the inbound email. (Planning an event of this scale is hard, I have empathy for anyone who does it lol.)
You get that this is a commute, right? Not charting a course through unfamiliar territory. A drive I've done thousands of times. I don't care if Waze thinks I should be on Harrison or Race. Fuck you, Waze. There's 4 reasonable routes; I just want to know which one.
It's an app with an audience of just me. That's the point. You're not supposed to want it. You have your own problems.
In my opinion, an OS that hosts a local agent can produce things for you on the fly in response to a particular task in addition to the baseline set of features.
Too often have I ran into "you can't do that" in operating systems, especially Macos where you either have to buy a paid app to implement what they should have already, or what you want is niche enough that you have to build it yourself.
Plus and agent can operate across your hardware including the cloud. I could ask something like "graphics cards are expensive and always out of stock so when some pop up can you notify me and show me the price etc" and then receive a rich alert on whatever device I happen to be using at the time it happens, once agents are aware of where we are and what we're doing (yes, it will be pretty dystopian with current governments and the drooling animals who vote for tribalism).
Or being able to be like "I want my photo gallery to show animated thumbnails when I've taken a series of photos of the same subject. Oh, little bit slower like one every second thanks" boom done.
And much UI will becomes defunct in the first place, UI only exists because of many annoying discrete steps we have to do to get something done. If I can just ask an agent to do the thing I want then I don't care about that.
It's like how I only need to make a PowerPoint now and then and every time I open it I've completely forgotten where all of the stupid fucking buttons are in the stupid fucking ribbon bullshit they made.
> plan mode has always been a prompt — it has never changed the toolset because doing so would break the prompt cache, and so would be expensive
seems like plan mode could turn off some tools, even if it doesn't change the set offered to the model, the ones that they have which would mutate your codebase could just not work with an error message, and plan mode could change permissions in the security approval prompt for "auto"
anyway, isnt the right way to know if plan mode helps or not, to run an experiment? we're all guessing unless we have data
read only agent mode sounds straightforward and useful to me
My friend's teenage son went through a phase where he would say a lot of awful shit that he didn't really understand. He explained to us why he and his friends thought it was funny, it was all the people freaking out about it.
People didn't just sit idly by and let bigotry make a comeback, there is a massive amount of pushback, to a degree that it can be unhinged. That is exactly why it has so much appeal to some people.
