wmf

7 ms·

181.
by wmf 2mo ago
Historically, selling to a small niche of price-insensitive customers (e.g. mainframes, defense) breeds complacency and eventual failure. You can see VMware go down that death spiral over its history where every release was more expensive w
182.
by wmf 2mo ago
No, because most ads are created by automated systems (many of which are scam/spam ads) and Meta doesn't want to lose revenue by slowing the system down to human speed.
183.
by wmf 2mo ago
I don't see why not. Somebody once wrote a simple Lambda that spawns a new instance when your spot instance dies so there's always at least one running.
184.
by wmf 2mo ago
They rebooted less than one year ago so this is decent progress. Obviously users don't care about progress though.
185.
by wmf 2mo ago
Clearly they're positioning it as a mid model.
186.
by wmf 2mo ago
He's failing up and out of the blast zone. He won't be responsible if Gemini 4 fails.
187.
by wmf 2mo ago
That really doesn't sound like a good business though.
188.
by wmf 2mo ago
More like the rack isn't viable for SaaS.
189.
by wmf 2mo ago
Since Guardian Angel is an inference vendor [1] hopefully they will set sustainable prices from day one so they don't need to increase prices later. [1] The announcement explains that they can't use APIs for privacy reasons.
190.
by wmf 2mo ago
Now that is a vague headline. Is the secret ingredient crime?
191.
by wmf 2mo ago
2026: $1,000/month 2027: $100/month 2028: $10/month
192.
by wmf 2mo ago
SQL supports quoting but LLMs do not. This makes it much more difficult to block prompt injection.
193.
by wmf 2mo ago
What can't be replicated is a culture of simplicity, quality, and security. I've seen this firsthand working at IBM.
194.
by wmf 2mo ago
No, rumors say 5-10T.
195.
by wmf 2mo ago
You can do that (tinygrad style) if you want.
196.
by wmf 2mo ago
Yes, ROCm can be used to run frontier models and is being used by OpenAI, Anthropic, and Meta.
197.
by wmf 2mo ago
https://github.com/antirez/ds4/pull/484 I assume this is parallel work.
198.
by wmf 2mo ago
The company is liable because the employee was acting on behalf of the company.
199.
by wmf 2mo ago
No, they believe all the data centers will be needed and will operate profitably.
200.
by wmf 2mo ago
The imminent third-party benchmarks will cover that.
201.
by wmf 2mo ago
Fable 5.1 is coming, then GPT-6...
202.
by wmf 2mo ago
The article says he did contact legal and I guess they declined to intervene.
203.
by wmf 2mo ago
Given the memory-unsafe vice signaling, maybe this can become the official terminal of Fox News and the Department of War.
204.
by wmf 2mo ago
Yeah, it's almost always better to settle than to go to court. But he's going to get less than he's asking for. And I bet he really wants an apology which he won't get.
205.
by wmf 2mo ago
There's infinite work to be done, so higher productivity makes people worth more. (Obviously this doesn't apply if AI can do everything but we're not there yet.)
206.
by wmf 2mo ago
This guy is going to lose. If a company is violating no laws and your manager tells you to drop it, you can either drop it or quit. You don't get to choose how to interpret company policy.
207.
by wmf 2mo ago
Obviously not. And if you look at the things LLMs do poorly there's clearly plenty of room for improvement.
208.
by wmf 2mo ago
Their solution is to "force open" iOS and Android through the DMA, not to create competition which they know won't work.
209.
by wmf 2mo ago
I assume this has head of line blocking. I wonder if evil networks would allow fake TCP through.
210.
by wmf 2mo ago
This thread is just as low quality and it's also a dupe.

More ›