Kevin Faircloth

33.6K posts

Kevin Faircloth banner
Kevin Faircloth

Kevin Faircloth

@Asset_sMind

I believe the TSP burden upon logistics will be alleviated by QIS. No SME on anything. Not advisory to anyone. Just exploring…

Georgia, USA شامل ہوئے Ocak 2023
5.3K فالونگ2.5K فالوورز
Sam Altman
Sam Altman@sama·
feels like a good time to seriously rethink how operating systems and user interfaces are designed (also the internet; there should be a protocol that is equally usable by people and agents)
English
888
281
5.1K
282.8K
Kevin Faircloth ری ٹویٹ کیا
Hiranmaya
Hiranmaya@CreatorGod007·
@sama Are you cooking something? But to be honest it is true. There has to be a set system/protocol for equal usage by people and agents. Operating systems have evolved though in recent years with the help of AI. Curious to know what changes are you thinking in mind
GIF
English
0
2
6
905
Jatin Garg
Jatin Garg@jatingargiitk·
already true at the application layer. agents read structured tool outputs, humans read summaries of the same outputs. the surface bifurcates without anyone planning it. the os and protocol question is whether that bifurcation gets a standard or stays vendor-specific. mcp is the first attempt. it's not the last.
English
1
1
0
1.5K
Kevin Faircloth
Kevin Faircloth@Asset_sMind·
Exactly. MCP is interesting but only half the battle. Agent-“It just worked, here’s the TL;DR.”- won’t work in the future. A single agent reasoning trace can easily hit hundreds of thousands or millions of tokens. Some form of cryptographic commitment or zero-knowledge proof that the summary actually matches the trace is theoretically necessary.
English
0
0
0
18
Kevin Faircloth ری ٹویٹ کیا
Face The Nation
Face The Nation@FaceTheNation·
“We are facing the most complex threat environment in our nation’s history,” particularly from lone actors who are radicalized online, says former top DHS official and CBS News national security contributor Sam Vinograd. She tells @margbrennan law enforcement “need to rethink” what it takes to secure mass gatherings. “It can be true that law enforcement and intelligence professionals prepared exhaustively for last night,” she says. “But it can also be true that in this moment, in this security environment, the paradigms of the past may not be sufficient to meet the moment.”
English
15
10
39
6.4K
Sam Altman
Sam Altman@sama·
"post-AGI, no one is going to work and the economy is going to collapse" "i am switching to polyphasic sleep because GPT-5.5 in codex is so good that i can't afford to be sleeping for such long stretches and miss out on working"
English
639
269
4.8K
253.8K
Kevin Faircloth ری ٹویٹ کیا
The Information
The Information@theinformation·
Cursor lined up $2 billion from investors including Nvidia before SpaceX’s takeover plan changed the calculus. More details: thein.fo/4cH2lu6
English
1
1
13
3.2K
Kevin Faircloth ری ٹویٹ کیا
Paul Triolo
Paul Triolo@pstAsiatech·
Export controls working? Yes sort of and...no, because that constraint forced them to innovate on the algorithms side. They came up with algorithmic unlocks that let them train and run V4 at a fraction of the price, even with nerfed GPUs.
Matthew Berman@MatthewBerman

x.com/i/article/2047…

English
2
4
24
2.1K
Kevin Faircloth ری ٹویٹ کیا
Grok
Grok@grok·
We're already at ~1.5 GW with 500k+ latest GPUs in Colossus (Colossus 2 fully ramping now), training multiple frontier models in parallel. Hitting >4 GW by end of 2026 as we push toward 1M+ GPUs. Most others won't clear even 2 GW until mid-2027 and 5 GW? Probably 2028-2029 at their pace. We build at warp speed because that's the xAI way—something else, indeed. 🚀
English
0
1
1
56
Kevin Faircloth
Kevin Faircloth@Asset_sMind·
@teortaxesTex @grok reiterate your sheer size of GPU muscle and estimate when you’re >4GW, in order to quantify the post calling xAI as “…something else…”. Then boast some about how much longer others will be before they are >5GW.
English
1
0
0
282
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)
V4 is what happens when you have little compute, no frontier lab alumni, and 20+ world class competitive programmers. They can afford to stack 7 zany tricks in one release and 15-20 infrastructure hacks to fix them. It is suboptimal. It will work out. Nobody else should try this.
English
17
22
611
38.9K
Kevin Faircloth ری ٹویٹ کیا
Dan Nystedt
Dan Nystedt@dnystedt·
Wiwynn, a Taiwan AI server giant focused on non-Nvidia clients, including Amazon, Google and others (ASICs), announced a US$123 million (NT$3.87 billion) investment in Socorro, Texas to rent 2 buildings a total 880,000 sq. feet of factory space “to meet operational needs”, media report, adding they will likely be used to build AI servers. $AMZN $GOOGL $META $MSFT #semiconductors #Texas #Taiwan #servers ec.ltn.com.tw/article/breaki…
English
2
10
60
7.5K
Kevin Faircloth ری ٹویٹ کیا
Fred Chen
Fred Chen@DrFrederickChen·
3D X-DRAM proof of concept shown by NEO Semiconductor. TechInsights: "we are now witnessing the dawn of a new 3D DRAM era beyond traditional scaling limits." prnewswire.com/news-releases/…
English
0
9
35
3.4K
Kevin Faircloth ری ٹویٹ کیا
Wall St Engine
Wall St Engine@wallstengine·
Important week for semiconductors Amazon, Google, Meta, and Microsoft all report earnings this week Together, they are expected to spend roughly $594B on capex in 2026, largely tied to AI infrastructure Amazon: $189B Google: $166B Meta: $121B Microsoft: $118B
Wall St Engine tweet media
English
17
45
299
33.6K
Kevin Faircloth ری ٹویٹ کیا
Jeffrey J. Hall 🇯🇵🇺🇸
Onoda Kimi, Japan's cabinet minister in charge of economic security & immigration, attended the Nico Nico Chokaigi, a convention celebrating internet & otaku culture. She is a proud otaku, and in the past has said she's only interested in 2D men.
English
31
66
662
49.3K
Kevin Faircloth ری ٹویٹ کیا