Lisan al Gaib's banner

Lisan al Gaib

@scaling01 • 38,297 subscribers

lead them to paradise https://t.co/IiP4VZlGU3

Shorts

Andrej Karpathy calls AI Agents slop "Overall, the models they are not there. And I feel like the industry [...] it's making too big of a jump and it's trying to pretend that this is amazing. And it's not—it's slop! And I think they are not coming to terms with it. And maybe they are trying to fundraise or something like that, I'm not sure what's going on."

Andrej Karpathy calls AI Agents slop "Overall, the models they are not there. And I feel like the industry [...] it's making too big of a jump and it's trying to pretend that this is amazing. And it's not—it's slop! And I think they are not coming to terms with it. And maybe they are trying to fundraise or something like that, I'm not sure what's going on."

3,099,121 Aufrufe

Sora-2 counting the R's

Sora-2 counting the R's

310,483 Aufrufe

Sora-2 brainrot: a cat walking on the ceiling as if it were the floor

Sora-2 brainrot: a cat walking on the ceiling as if it were the floor

306,977 Aufrufe

Stargate Abilene - overview, progress and speculation + satellite pics Power: 1.2 GW total for 8 buildings ~44,600 m² (~8+ American football fields)., 4 halls each Timeline: Construction began June 2024 (Phase 1: 2 buildings). June 2025 the first two buildings were completed and the campus began early training/inference. Phase 2 adds 6 more buildings; full campus mid-2026 GPU count (planned): March 2025: ~400k GB200 chips (~50k/building). Oct 2025 (Larry Ellison): >450k GPUs (>56.25k/building) Cooling approach: Direct-to-chip, closed-loop, zero-evaporation liquid cooling. Water: ~1,000,000 gallon one-time fill per building; ~12,625 gallons/year per building for maintenance and quality Chillers: 80 per building. Building 1 in red: smaller frames (something like Vertiv LVZ170), up to ~2.7 MW each. Newer buildings in green: 20% larger frames (something like Vertiv LVZ230), up to ~3.1 MW each. Usually around 2 MW/unit and only 50-60 during normal operation. The rest are for backup or extreme heat. -> likely increased cooling capacity to accomodate 50k more GPUs over 8 buildings On-site generation: Parker Hannifin supplying 29 dual-fuel GE Vernova LM2500XPRESS turbines (~35 MW each) for ~1.015 GW on-site capacity. Currently 9 deployed at 315 MW with 10th one in progress Speculation: - Compute: 196 racks/hall → 784 racks/building → 56,448 GPUs/building → 451,584 GPUs campus-wide - Power: 784 × 132 kW = 103.488 MW IT/building → at PUE 1.15 ≈ 119 MW/building → ~952 MW campus total (leaving headroom under 1.2 GW) - Site status as of October 30th 2025: Construction started for all buildings in phase 2 and a third building appears nearly complete with chillers installed

Stargate Abilene - overview, progress and speculation + satellite pics Power: 1.2 GW total for 8 buildings ~44,600 m² (~8+ American football fields)., 4 halls each Timeline: Construction began June 2024 (Phase 1: 2 buildings). June 2025 the first two buildings were completed and the campus began early training/inference. Phase 2 adds 6 more buildings; full campus mid-2026 GPU count (planned): March 2025: ~400k GB200 chips (~50k/building). Oct 2025 (Larry Ellison): >450k GPUs (>56.25k/building) Cooling approach: Direct-to-chip, closed-loop, zero-evaporation liquid cooling. Water: ~1,000,000 gallon one-time fill per building; ~12,625 gallons/year per building for maintenance and quality Chillers: 80 per building. Building 1 in red: smaller frames (something like Vertiv LVZ170), up to ~2.7 MW each. Newer buildings in green: 20% larger frames (something like Vertiv LVZ230), up to ~3.1 MW each. Usually around 2 MW/unit and only 50-60 during normal operation. The rest are for backup or extreme heat. -> likely increased cooling capacity to accomodate 50k more GPUs over 8 buildings On-site generation: Parker Hannifin supplying 29 dual-fuel GE Vernova LM2500XPRESS turbines (~35 MW each) for ~1.015 GW on-site capacity. Currently 9 deployed at 315 MW with 10th one in progress Speculation: - Compute: 196 racks/hall → 784 racks/building → 56,448 GPUs/building → 451,584 GPUs campus-wide - Power: 784 × 132 kW = 103.488 MW IT/building → at PUE 1.15 ≈ 119 MW/building → ~952 MW campus total (leaving headroom under 1.2 GW) - Site status as of October 30th 2025: Construction started for all buildings in phase 2 and a third building appears nearly complete with chillers installed

143,745 Aufrufe

Videos

Anya Rossi

sweetdream.ai

SweetDream.ai•Sponsored•Livecam

Watch Anya Live

Anya is streaming live right now! Join her private show and enjoy exclusive content.

Exclusive private shows

1.2k viewers online

Private Show

Join now for exclusive access

Free preview available • Premium content

GPT-5.4 is special LisanBench: GPT-5.4 vs Opus 4.6 vs Gemini 3.1 Pro

GPT-5.4 is special LisanBench: GPT-5.4 vs Opus 4.6 vs Gemini 3.1 Pro

265,203 Aufrufe • vor 3 Monaten

GPT-5 vs Grok-4 exact same prompts, wildly different output in web lmarena

GPT-5 vs Grok-4 exact same prompts, wildly different output in web lmarena

566,992 Aufrufe • vor 10 Monaten

Dario Amodei, CEO of Anthropic says that researcher-lead AI companies, like Anthropic and Google will succeed going forward. He declined to comment about what would happen to companies, like OpenAI, xAI and META, that aren't led by researchers.

Dario Amodei, CEO of Anthropic says that researcher-lead AI companies, like Anthropic and Google will succeed going forward. He declined to comment about what would happen to companies, like OpenAI, xAI and META, that aren't led by researchers.

180,189 Aufrufe • vor 4 Monaten

HOLY FUCKING SHIT - WATCH THE WHOLE VIDEO I asked Gemini 2.0 what the meaning of life is but it is only allowed to use images. OpenAI made a generational fumble by not releasing GPT-4o fully multimodal Fully multimodal models are the weirdest and scariest things in existence wtf this is addicting

HOLY FUCKING SHIT - WATCH THE WHOLE VIDEO I asked Gemini 2.0 what the meaning of life is but it is only allowed to use images. OpenAI made a generational fumble by not releasing GPT-4o fully multimodal Fully multimodal models are the weirdest and scariest things in existence wtf this is addicting

370,769 Aufrufe • vor 1 Jahr

Microsoft has access to ALL of the hardware accelerator IP from OpenAI, which is why Satya is not worried about their in-house Maia accelerators not progressing as fast as Google TPUs

Microsoft has access to ALL of the hardware accelerator IP from OpenAI, which is why Satya is not worried about their in-house Maia accelerators not progressing as fast as Google TPUs

156,886 Aufrufe • vor 6 Monaten

This is where we are with Claude Mythos Governments are waking up and talks about nationalization will be happening within a year

This is where we are with Claude Mythos Governments are waking up and talks about nationalization will be happening within a year

29,946 Aufrufe • vor 1 Monat

and so it begins

and so it begins

25,280 Aufrufe • vor 3 Monaten

$this is now a much more accurate representation of the space and it still holds that GPT-5.4 is a bit more explorative it's now based on kamada–kawai layout instead of the default spring layout, and there are much more nodes and edges than before (still need to optimize it tho) but it's not close to all nodes or edges in the neighborhood, still only a tiny fraction the fuzzball is very dense$

this is now a much more accurate representation of the space and it still holds that GPT-5.4 is a bit more explorative it's now based on kamada–kawai layout instead of the default spring layout, and there are much more nodes and edges than before (still need to optimize it tho) but it's not close to all nodes or edges in the neighborhood, still only a tiny fraction the fuzzball is very dense

19,040 Aufrufe • vor 2 Monaten

I asked Claude 4 Sonnet to build Flappy Bird. It took longer to screen record than for it to code it. it literally one shot this in pure HTML, CSS, JS

I asked Claude 4 Sonnet to build Flappy Bird. It took longer to screen record than for it to code it. it literally one shot this in pure HTML, CSS, JS

48,923 Aufrufe • vor 1 Jahr

I can't stop playing with this tool that GPT-5 made it's so mesmerizing

I can't stop playing with this tool that GPT-5 made it's so mesmerizing

41,221 Aufrufe • vor 10 Monaten

LMARENA EVOLUTION OF TOP 10 MODELS BY ELO OVER TIME Since November 20th, only Google and OpenAI models have been in the Top 10 !!! Anthropic hasn't made the Top 10 since September 27th.

LMARENA EVOLUTION OF TOP 10 MODELS BY ELO OVER TIME Since November 20th, only Google and OpenAI models have been in the Top 10 !!! Anthropic hasn't made the Top 10 since September 27th.

44,424 Aufrufe • vor 1 Jahr

Keine weiteren Inhalte verfügbar