So far I have mixed impressions, but they do indeed seem noticeably weaker than comparably-sized Qwen3 / GLM4.5 models. Part of the reason may be that the oai models do appear to be much more lobotomized than their Chinese counterparts (which are surprisingly uncensored). There's research showing that "aligning" a model makes it dumber.
The censorship here in China is only about public discussions / spaces. You cannot like have a website telling you about the crimes of the party. But downloading some compressed matrix re-spouting the said crimes, nobody gives a damn.
We seem to censor organized large scale complaints and viral mind virii, but we never quite forbid people at home to read some generated knowledge from an obscure hard to use software.
That is fine as long as the input / output is always in UTC... but at the end of the day you often want to communicate that timepoint to a human user (e.g. an appointment time, the time at which some event happened, etc.), which is when our stupid monkey brains expect the ascii string you are showing us to actually make sense in our specific locale (including all of the warts each of those particular timezones have, including leap second, DST, etc.)
It is not, if what the user expects to stay constant is their local calendar/wall clock time rather than the UTC instant. Which is usually the case. This is a transformation that needs late binding as DST and timezone rules can change, so it can't just be handled as a localisation transformation on input/output
Yeah, I don't understand these sort of hot takes in the articles like it. As said middle-management dimwit, isn't sort of the definition of a high performer is that they perform without much coaching? That doesn't mean lack of management as the article implies - it means setting them up for success.
Set them on a task, give them a few constraints as needed, and act as a bulldozer to remove blockers as they come to you with them. That's roughly the management a high performer needs in my book. Make sure you listen to them when they complain about something, and fix it within reason. Or explain why you cannot. Don't let penny wise pound foolish stuff fester. Advocate for merit based raises and bonuses - they are making you look good so they should be your first priority when expending political capital. Make sure higher ups and the rest of the org knows about their accomplishments. Simple stuff!
IMHO, the overhead of perpetually babysitting compiler diagnostics or performance metrics to ensure your latest update didn't confound the auto-vectorizer is never a net positive over just using something like xsimd, Google highway, etc.
Agree 100%... Getting caught using a phone while driving, should be punishable by a suspended 5-year prison sentence contingent upon completion of a 1 year smartphone ban. Get caught using anything other than a flip phone at any point during the next year and you have to serve the prison term.
I think no smartphone ban is a little harsh. It's needed for modern banking or interacting with government services or for job hunting or even work. Even the government hands out free smartphones via the Lifeline / "ObamaPhone" program
There's a PC alternative for all those. And even a computer free one.. at least in my EU country. To me it would not be a bad idea, and the governements should ensure it's still possible to do all those things without a smartphone.
> thief breaking into your house, stealing all. . .
This is where your analogy is flawed. You are pre-supposing the "defendant" is indeed the thief that stole your property. Whereas that is entirely a legal determination which is the outcome of a trial AND at the heart of this discovery request. More aptly if you thought steve stole your red Ryder bb gun, and Steve was indeed found to be in possession of a red Ryder bb gun, it would still be the prosecution's burden to prove that Steve stole it from you (instead of purchased it from a store).
Similarly here, if NYTimes is claiming that openai's gpt4 illegally reproduces "to be or not to be. . ." (Or whatevs) from issue #8628 page 76, it's still their burden to prove that is actually a thing that is both copyrightable and that they own the copyright to vs. openai just reproducing Hamlet instead of a nytime's reporter's particular review of a production of hamlet in that issue. Etc. etc.
More germanely, if you point an llm at a pile of source documents and ask it to write a newspaper article, it'll happily do so in 2024. Understanding if/how this is fundamentally different from what a reporter does when synthesizing that same article goes to the very heart of this case (i.e. which transformative works are indeed copyrightable)
It's not an exact analogy, but I think it captures the basic moral point of what's at issue here.
It's perfectly obvious what OpenAI did - and it knows perfectly well how limited its horizons will be if it can't build its empire on the basis of stolen material.
That's why it's now hissing and spitting like a cornered animal.
So far I have mixed impressions, but they do indeed seem noticeably weaker than comparably-sized Qwen3 / GLM4.5 models. Part of the reason may be that the oai models do appear to be much more lobotomized than their Chinese counterparts (which are surprisingly uncensored). There's research showing that "aligning" a model makes it dumber.