ML (@matvl77) profile picture

Founder/CEO at https://teapawt.org/t.co/7kwgfnawzU QF researcher in Amsterdam Art enjoyer

Amsterdam, The Netherlands
Joined March 2023
Photo by ML (@matvl77)Photo by ML (@matvl77)Photo by ML (@matvl77)Photo by ML (@matvl77)Photo by ML (@matvl77)Photo by ML (@matvl77)Photo by ML (@matvl77)Photo by ML (@matvl77)Photo by ML (@matvl77)Photo by ML (@matvl77)
An interesting reality right now is that the best way to use Sol is through Opus main thread in CC. Let a smarter model answer GPT's stupid questions.
27
Normies are truly a manifestation of the last man. Existing in blis and arrogant ignorance of everything around them.
1
18
Rtx pro 6000 are more valuable then silver by gram. Including cooling.
30
Everyone kept hyping "open source is x months away" until main labs needed an industry wide mog at every price/ intelligence level. Welcome back total Anthropic victory.
1
2
59
You can do Fourier transforms in O(N log(N)^0.9999999999999) time. Fuck this gay chud universe
91
245
111
7,141
1,399,673
In the battle between OAI and Anthropic, the consumer can choose between slow models/cut limits or 0-bit Claude.
1
33
Tsk tsk, only I make a joke about Google beeing out of the race, they decide to mog the industry. Would be great if it is NOT like other Geminis though
40
Got 65k worth of OAI "credits". It sounds like a lot, but if you run the numbers it's 1 or 2 weeks of usage. Literally cut the limits in half and give you 1-2weeks extra as consolation.
31
Sol6 is such a disappointment, 5.6 was the first model you can confidently give business tasks and return to the finished product. Sol6 is like a lobotomized version + worst of opus yapping and not checking your intent.
33
Anthropic watermarks on text are insane. I've made a mistake asking Claude to formalize a paper from my ideas + rough draft and I got the most jarring piece of slop I've seen in a while. Just absolutely impossible to read. Reminded me of the "translate Chinese to English 20 times" memes from early google translate, absolutely bizarre choice of language, and sentences that don't flow together at all. I fought for a bit but I just went insane and it was normal Claudeish, but no. Even gpt spotted that something wasn't right.
2
1
66
I don't understand how they don't see it. One of the biggest capability regressions I've seen to date. It's basically dead for writing.
30
I hate normy housing corporations. Why would you limit the internet speeds if you have fiber to every unit??? Let me pay extra. I don't care; just give me gigabit speeds. Real convo btw: > I see you have a fiber switch in the building; make my port 1gig. > No sir, the cables can only do 100mbit > It's fiber; I'm quite sure it's an artificial limit > Well yes, you are right, but if we increase it for everyone, the network will get overloaded > I don't care, give me a dynamic link that would scale if there is capacity. Its litereally one button on your end. > No sir, we can't do it; it's a physical limitation.
122
It's quite interesting to see a lot of Chinese companies pushing the envelope on consumer tech. I distinctly remember how just 5 years ago, to get the same features you would need $2-3k of hardware, and now you can get it for a hundo with free shipping.
49
>@nikitabier steps down as X head of product. > Android app becomes unusable garbage 2 weeks later Did he run literally everything???
64
"Jarred's input was mostly limited to sending Claude messages of encouragement (mostly variants of "keep going" or "believe in yourself"). This seems to have helped Claude overcome some initial skepticism that it could make meaningful progress." Once who are the most surprised by the capabilities of the new models - are the models themselves. Eventually we will get models that will fully understand their own capability and that would be an ASI moment.
We asked an unreleased research version of Claude to take a stab at the Riemann hypothesis. It didn’t solve it, but it did make strides on a related problem: it increased the lower bound for the fraction of zeros of the Riemann zeta function that satisfy the hypothesis from 41.6% to 67.2%. anthropic.com/research/riema…
133
Claude discovered the Gem vs. Coal dichotomy by reviewing scheduled LinkedIn posts. The models become funnier and funnier.
🤖 Made with AI
37
It's funny how GPT Luna is really useful (but only with Opus as an orchestrator). GPT Sol doesn't delegate tasks well. Opus treats Luna runners with extreme disdain. You can feel the hate and disrespect, but it will run 100-1000s of them without a second thought.
27
I bought a print calibrator and now I'm stuck between Japanese and German designed software . . . Is this what hell feels like ?
34
Sometimes you get a real world comparison so wild you have to stop and check. 20-30x grain productivity per hectare. Even medieval charlatans would find this ratio exadurated.
Cutting wheat yesterday: roughly 100 grains harvested for each grain sown. In the 1300s on the Bishop of Winchester's manors, 5:1 was a good harvest, 3:1 a bad one. Thank you, Haber-Bosch nitrogen fixation and hydrocarbons, for feeding billions and sparing land.
1
87