Y
HN Search
Hacker News Search
new
|
comments
|
top
|
jobs
Topfi
searching PlanetScale…
1.
▲
2.
▲
3.
▲
4.
▲
5.
▲
6.
▲
47 ms
·
751.
▲
by
Topfi
3y ago
I have long had a theory for why Stadia handled CP77 so well compared to the other platforms. CDPRs management of CP77s development was, to put it mildly, less than ideal and any extensive period of crunch will always come at the detriment
752.
▲
by
Topfi
3y ago
I am sure most at OpenAI realize that they don't have a moat[0] that will keep them ahead for an extended period of time. Being a first mover does have its advantages of course, funding and resources from MSFT also help, but if they c
753.
▲
by
Topfi
3y ago
>>It’s unethical to use a person’s name to publish a study or charge for co-authorship. Maybe I am alone on this, but is "collaborating" on topics one has little to no in-depth knowledge in and essentially just checking the
754.
▲
by
Topfi
3y ago
It is very possible that they did not predict this specific down stream effect, for one of many reasons. I do not want to discount that, though, as I am not at Twitter and am not on a first name basis with anyone there. Maybe there are some
755.
▲
by
Topfi
3y ago
That is a good question, but on the two specific occasions when something of this sort did not make me leave instantly, there were very specific factors at play. I cannot comment on the first situation as that would require an extensive pos
756.
▲
by
Topfi
3y ago
Yes, that is very much true. Even if, as some have commented, the people remaining at Twitter lack oversight over the entire codebase to predict that this specific change could have such a knock-on effect, not having testing and staging env
757.
▲
by
Topfi
3y ago
You are correct that Fedora is mainly community driven and currently receives a large part of its funding from Red Hat. I could have phrased my initial post better, I am currently not concerned that their actions could negatively affect the
758.
▲
by
Topfi
3y ago
Speaking from very painful, personal experience, few things are more agitating than being forced to execute on something you fully know is a horrible idea, especially when you tried and failed to communicate this fact to the individual push
759.
▲
by
Topfi
3y ago
My recommendation would be Fedora Silverblue[0]. I have for years used different flavors of Ubuntu and Arch, but must admit, I always had some issues, no matter the distro. Mind you, purely counting the problems, the amount was either compa
760.
▲
by
Topfi
3y ago
For me, this loops back to the age-old question "Is there honor among thieves?". Or, to put it in another way, whose to say that they deliver at all? The site appears barely functional, their own social media accounts have laughab
761.
▲
by
Topfi
3y ago
I honestly really like this, both conceptually and from an implementations point of view. Assisting in the creation of single purpose functions is, in my experience, by far the most reliable use case for LLMs in regard to code creation at t
762.
▲
by
Topfi
3y ago
Whilst that is true, I feel that if that human is an educator employed by a reputable institution, answering questions in their field, the expectation of most is going to be that they provide accurate information. As this is an introductory
763.
▲
by
Topfi
3y ago
Part of that is, in my eyes, that it is an apt comparison. Some of the things that ended up making a touch only interface work well on the iPhone (I used Windows Mobile 6.0 with only a touchscreen at the time, so I am speaking from experien
764.
▲
by
Topfi
3y ago
A great choice, my favorite for daily wear, but Anthelios is basically sold everywhere over here. I have no dog in this fight, especially because as someone who often finds themselves defending EU regulations to people who grew up in the St
765.
▲
by
Topfi
3y ago
This honestly seems like a reasonable, sober approach to LLMs that considers how they may serve journalists in a few key ways, whilst retaining them as the ones actually writing the content. Using LLMs to help sift through large datasets an
766.
▲
by
Topfi
3y ago
You are absolutely right for the moment, but the question is whether this will always remain the case. I am honestly doubtful and fairly certain this will be more a question of when than if. Seeing as both Nvidia and Apple continue making c
767.
▲
by
Topfi
3y ago
Regulation in the way OpenAI appears to campaign for in this response is, in my eyes, going to have a very similar trajectory to the consistent and misguided attempts at enforcing backdoors in or a complete ban of modern encryption. Attempt
768.
▲
by
Topfi
3y ago
Discussion of the multifaceted issues causing inflation across the globe (even in countries with a less severe increase of their money supply across the same timeline) aside, does this in any way negate their position on Tesla operating the
769.
▲
by
Topfi
3y ago
This was an international issue that affected me personally here in Austria. Tried paying and received a cryptic error. Withdrawals via ATM, even at the bank I was affiliated with, yielded similar results. As I encountered this just after c
770.
▲
Banks solve problems with Mastercard payments
(independent.co.uk)
6 points
by
Topfi
3y ago
|
2 comments
771.
▲
by
Topfi
3y ago
Over a year ago, I switched to an M1 Max 14-inch as my main computer from an AMD Renoir equipped laptop running Fedora Silverblue with Pantheon. Previously used MacOS from Snow Leopard to Yosemite and very much liked it back then. Everyone
772.
▲
by
Topfi
3y ago
Fedora Workstation and especially Silverblue made Linux work for me in a way no distribution I have used before has been able to. Being bleeding edge yet stable and reliable just made it so pleasant, coming from a mix of Slitaz, Elementary
773.
▲
by
Topfi
3y ago
I can see this going only one way.[0] [0] https://www.reddit.com/r/ListOfComments/wiki/downvoted/
774.
▲
by
Topfi
3y ago
Maybe I am missing something, but the Framework focused posts I have seen seem to mostly come from long active accounts by actual community members. Think this is more simply a case of the product being interesting to the people on here. Fr
775.
▲
by
Topfi
3y ago
Honestly, at least when it comes to code in widely used languages, GPT-4 is very good at catching mistakes in generated output after one makes a simple request for a second check. In most cases, there isn't even a need to explain what
776.
▲
by
Topfi
3y ago
That is a fair point, regulation often gets watered down as it undergoes revision after revision, though in fairness, most cases can not be solely blamed on US tech. The stricter timetable for an ICE ban, to use something mentioned, was mai
777.
▲
by
Topfi
3y ago
salesforce really seems to invest a lot of resources into coding focused applications for LLMs, which is of course great, especially as they seem very transparent, sharing both papers and usable implementations[0]. However, I feel that I am
778.
▲
by
Topfi
3y ago
Thanks to the entire Asahi team, your work is truly incredible and so far beyond my pay grade that words fail me. I honestly recently tried and very much struggled to communicate why I was so amazed by your project when talking with friends
779.
▲
by
Topfi
3y ago
>Break down a big change and ask for a series of smaller, self contained steps. These are all good tips and very much align with my experience over the months, but breaking a complex task into as small, self-contained and logical steps a
780.
▲
by
Topfi
3y ago
That's a great development. Reducing vendor lock-in removes one of the bigger reservations some mentioned in regards to WebAuthn.
More ›