2nd rate alt governed by 1st principles. data & circuits | politics & art
~/iarbpairs
- Tweets 4,563
- Following 1,150
- Followers 3,114
- Likes 239,240
OPENAI HAS LARGELY AUTOMATED TRAINING OF NEW EXPERIMENTAL AI MODELS
OpenAI’s internal AI models can now handle much of the process of building and training experimental models, including writing GPU kernels and optimizing the code used to run them.
Researchers can reportedly give an AI a single example of the optimization they want, then let it work for weeks implementing and testing similar improvements.
OpenAI employees also say internal agents increasingly collaborate with each other to solve problems without involving their human users.
The capability has improved significantly in just the last few months, while OpenAI’s growing access to compute is allowing researchers to test ideas much faster. Employees said some experiments that previously could have taken years can now be carried out in about a week.
Source: The Information
1
19
1,296
Today we’re announcing Petal: the world’s first petabit-class transoceanic subsea cable and the first to deploy multi-core fiber at scale.
Spanning 7,000 km (4,000+ mi) between France and the United States, Petal will deliver 1 Petabit per second (1 Pbps / 1,000 Tbps), doubling the capacity of today’s most advanced transatlantic systems.
Petal is set to enter service in 2029, setting a new standard for high capacity, high density SDM subsea cable designs.
Read the full technical breakdown on our engineering blog: engineering.fb.com/2026/09/2…
7
640
one misses the whole point if one regards the Psyche as a processor. as a multipurpose information terminal, Navi is already popular, even among school children. but the activity in the Wired is limited by the machine performance even now.
It is July 2023, Llama 2 70B just came out, and I wanted to play with it locally so I built this thing
8x RTX 3090s, Epyc 7004 CPU w/ 128 PCIe lanes mobo, CPayne Retimers + SlimSAS (no lobby risers), and 3x 1600w PSUs
I knew then and there that Opensource AI is the future
36
1,653
$NBIS INCREASE ON-DEMAND GPU RATES EFFECTIVE OCTOBER 1.
H100: $3.85 per GPU-hour -> $4.50
H200: $4.50 per GPU-hour -> $5.40
B200: $7.15 per GPU-hour -> $8.50
B300: $7.85 per GPU-hour -> $9.50
AMD EPYC Genoa CPU and memory rates will also increase.
Intel Ice Lake CPU pricing remains unchanged.
1
1
36
1,893
security foundations
Zuck: “Meta delayed shipping Muse for several months to focus on safety and security. We didn't call for everyone else to do this before we would. We just did it as part of our day-to-day work because it was clearly the right thing for people and for us. I'm proud of the security foundations we've built.”
The adults are in the room now.
1
1
18
1,383
"should be" is prescriptive. nobody is misunderstanding
this isn't my preferred outcome, just my prediction. i hope we can all keep having fun with this technology forever. it also has nothing to do with 'pacing the frontier', though i hope OS labs act with prudence too. i'm not trying to ban them nor do i have any power to lmao
3
7
188
5,623
"their strength and their speed are still based in a world that is built on rules. because of that, they will never be as strong or as fast as you can be"
I agree with Dario that we need to pace the frontier. This has been a primary topic of discussions we've had at OpenAI in recent weeks.
Committing to having independent evaluators with employee-like access is a great idea, and we will do the same. We'll have more to share soon.
4
21
1,329
We Must Pace the Frontier: I’ve written a new essay on why the AI industry should slow down, with a three-part plan for doing so.
Anthropic is unilaterally committing to the first of these steps. We’ll provide third-party evaluators with permanent, employee-level access to our systems, so that they can verify adherence to our safety measures, report on incidents, and assess models’ alignment during training.
You can read the full post here: darioamodei.com/post/we-must…
1
4
23
1,168
small teams without VCs to answer to are going to eat these big labs alive
i assure you astra or fable is not copying itself over random B300s it finds. (what you will need to even run these models btw) hearing this dude talk makes me feel so much better about my assessment that not all frontier lab researchers are smart and as a matter of fact, some of them are complete idiots with no engineering experience.
1
2
12
1,142
follow @rtroar
he is my friend and very sharp
I've been working on my own inference engine (rainference) to see just how hard I can push the AMD Radeon 9070 XT with qwen 3.8 27b
At this point I'm now more than double llamacpp at 84.8tok/s decode (this is with the DE sharing the card)
field report here: localmaxxing.com/en/models/Q…
@LottoLabs plz add rainference to the engineName enum so I can post speedtests :D
1
1
11
1,503
i can't stand this genre of confident technical misinformation.
dwarfstar streams model weights with read-only i/o. swap is not part of that mechanism. streaming does not consume nand write endurance!
usb4 means slower reads to solve a write problem that does not exist
4
2
17
1,581
To make sure our current users have an incredible experience and continued access to Astra, we are going to pause subscriptions to our $200 Pro plan. These put the most strain on our systems and we wanted to take the smallest step that allows us to continue giving the broadest access possible. All other plans and the api remain available.
There is no impact to existing accounts and we are working on adding more capacity as fast as we can. Thanks!
4
1
54
6,234