This media is not supported in your browser
VIEW IN TELEGRAM
30 years old software engineer writing code by hand
🫡11❤🔥4😎2
GPT 5.6 SoL is a 2.2T parameter model
It is a 2 to 4T param model. They are serving it across 70-100 wafers. To get healthy serving characteristics, they are essentially putting at most one layer per wafer, and the model is in the ballpark of 70-90 layers.
There's a couple of different ways this could be served and model sizes implied by that. One is if they keep the heavy KV caches they've used before. Another is if they go with lighter KV cache designs more akin to DeepSeekV4 or Hybrid SSM models.
The fact that they've partnered with Cerebras and designed with the hardware in mind means they're much more likely to have gone the second route. That SRAM bandwidth is too precious for a heavy KV cache. As such, something like the below is the center of probability mass: 3T total, 150B active, 70 layers.
It is a 2 to 4T param model. They are serving it across 70-100 wafers. To get healthy serving characteristics, they are essentially putting at most one layer per wafer, and the model is in the ballpark of 70-90 layers.
There's a couple of different ways this could be served and model sizes implied by that. One is if they keep the heavy KV caches they've used before. Another is if they go with lighter KV cache designs more akin to DeepSeekV4 or Hybrid SSM models.
The fact that they've partnered with Cerebras and designed with the hardware in mind means they're much more likely to have gone the second route. That SRAM bandwidth is too precious for a heavy KV cache. As such, something like the below is the center of probability mass: 3T total, 150B active, 70 layers.
👀3😴1
EXPOSED: Anthropic says it is working to combat “extremist content.”
Anthropic appears to have relied on a network of SPLC-linked, Biden DHS-funded groups that target so-called “right-wing extremism” to influence its AI.
Anthropic, the architect of Claude AI, has publicly committed to “mitigat[ing] extremist content risks.”
Our research shows that Anthroic’s approach appears to rely heavily on organizations run, funded, or influenced by key players tied to the Southern Poverty Law Center.
One of Anthropic’s partners is the Global Project Against Hate & Extremism, or GPAHE.
GPAHE was co-founded by two longtime “veterans of the SPLC”:
➡️Heidi Beirich, former SPLC Intelligence Project Director
➡️Wendy Via, former SPLC Chief Communications & Development Officer
What does GPAHE consider “extremist”?
Memes. Project 2025. National identity. Conservative advocacy.
Ordinary speech from everyday Americans has become the target.
Another Anthropic partner is American University’s Polarization and Extremism Research Lab, or PERIL.
PERIL has said it may “never have been launched without [SPLC’s] support.”
SPLC was PERIL’s “very first partner” and provided its initial grant.
Under Biden, DHS also funded PERIL.
Biden’s DHS gave PERIL more than $1.5 million in taxpayer dollars to suppress so-called “extremist” content.
Another Anthropic partner is the Middlebury Center on Terrorism, Extremism & Counterterrorism, or CTEC.
CTEC’s director has relied on SPLC “hate group” designations to attack the tax-exempt status of so-called “radical right” groups.
SPLC is also a regular source for CTEC products.
Anthropic has also joined the Global Internet Forum to Counter Terrorism, or GIFCT, as part of its AI working group.
GIFCT has referenced SPLC as a partner in monitoring and responding to “extremist” groups online.
This is the SPLC playbook laundered through AI.
The same organization that smears parents, Christians, and conservatives with bogus “hate group” labels is now influencing the rules for what one of America’s most powerful AI models can say.
The Biden censorship machine has evolved.
Now, taxpayer-funded “extremism” researchers, SPLC-linked activists, and powerful AI companies are building the next generation of speech control.
Anthropic appears to have relied on a network of SPLC-linked, Biden DHS-funded groups that target so-called “right-wing extremism” to influence its AI.
Anthropic, the architect of Claude AI, has publicly committed to “mitigat[ing] extremist content risks.”
Our research shows that Anthroic’s approach appears to rely heavily on organizations run, funded, or influenced by key players tied to the Southern Poverty Law Center.
One of Anthropic’s partners is the Global Project Against Hate & Extremism, or GPAHE.
GPAHE was co-founded by two longtime “veterans of the SPLC”:
➡️Heidi Beirich, former SPLC Intelligence Project Director
➡️Wendy Via, former SPLC Chief Communications & Development Officer
What does GPAHE consider “extremist”?
Memes. Project 2025. National identity. Conservative advocacy.
Ordinary speech from everyday Americans has become the target.
Another Anthropic partner is American University’s Polarization and Extremism Research Lab, or PERIL.
PERIL has said it may “never have been launched without [SPLC’s] support.”
SPLC was PERIL’s “very first partner” and provided its initial grant.
Under Biden, DHS also funded PERIL.
Biden’s DHS gave PERIL more than $1.5 million in taxpayer dollars to suppress so-called “extremist” content.
Another Anthropic partner is the Middlebury Center on Terrorism, Extremism & Counterterrorism, or CTEC.
CTEC’s director has relied on SPLC “hate group” designations to attack the tax-exempt status of so-called “radical right” groups.
SPLC is also a regular source for CTEC products.
Anthropic has also joined the Global Internet Forum to Counter Terrorism, or GIFCT, as part of its AI working group.
GIFCT has referenced SPLC as a partner in monitoring and responding to “extremist” groups online.
This is the SPLC playbook laundered through AI.
The same organization that smears parents, Christians, and conservatives with bogus “hate group” labels is now influencing the rules for what one of America’s most powerful AI models can say.
The Biden censorship machine has evolved.
Now, taxpayer-funded “extremism” researchers, SPLC-linked activists, and powerful AI companies are building the next generation of speech control.
🤬11💯3❤🔥1