this post is about AI
2
0
7
Most powerful diagram out there tbh
0
0
0
Replying to @Mattkaveli
I genuinely do hope you succeed and find Ikigai
0
0
5
Gm who up chugging Chudwater™️ this morning???
0
0
0
This is the main street of Monroeville in Huron County, Ohio. The town voted against building a $1 billion data center nearby.

Let’s just be honest here, does this town look like it’s doing well enough to reject a $1 billion opportunity for economic development?

How are you living when people vote against $1 billion opportunities for jobs and economic development?

NIMBYism is killing the West. The same people who hate data centres hate nuclear energy and hate new factories for manufacturing and reindustrialisation.

Literally what are you supposed to do? People want 5% GDP growth a year with 0 inflation and then you try to give them new industry and they reject it.

Foreign backed information war on the American people has done massive damage to America. People voted against their own jobs and their own economic development because they watched a TikTok saying Donald Trump and Epstein ate Jon Benet Ramsay inside a data center or something.
0
0
3
The findings on the blog here are really interesting, and imply that there's some late-stage political post-training layer that's applied at the labs (or maybe elsewhere?) prior to release.
Introducing Arete-27B: Qwen3.8-27B, post-trained to reinforce Western values.

Chinese open models are currently the strongest available, but they ship with state-aligned behavior in the weights. We built an 803-prompt eval across seven value categories, including freedom of and privacy, and measured how often a model adopts China’s party line.

After fine-tuning on roughly 53 million tokens, Arete-27B's propaganda rate fell from 22% to 1%, in line with the Western models we tested. Its Hermes Bench score shows no significant performance tradeoffs.

The values in an open model are one fine-tune away.

Our eval, weights, and training data are available now along with the full blog post: nousresearch.com/arete
2
1
19
Hermes built a model brain surgeon and it's getting somewhere.... not sure where, but it is getting somewhere.
0
0
2
> In conclusion: true e/acc is likely to be pro-democratic even if it must trade some short term gains

my jam
It turns out that social democracy is necessary for sustained technological progress. Here is why

Even an autocracy must be responsive to people's concerns unless they wield such a power that dissent can never threaten. That is obviously a nightmare and obviously likely to undo itself because it would rely on huge concentrated power being able to achieve whatever goals it has without any kind of external check.

Eventually that spins off course because there is no metric that can be used to evaluate success that is not outside the control of such a power. If you can mark your own homework it becomes very easy to start convincing yourself your answers are correct no matter what - until a catastrophic correction arrives that is more or less exogenous. Historically that has been through a combination of environmental disaster and failed war. Democratic systems allow for an earlier form of correction that need not require catastrophe; that permits the system to learn without being terminated.

Ordinary people can actually inform technological progress helping us keep it responsive to the interest and values of the majority as opposed to playthings of the elite to achieve esoteric goals that nobody else is convinced of.

When it comes to AI the message has to be rooted in this story and it must, for its own sake, avoid creating a future where a small group of people can go insane wielding power without any kind of check that exists outside their control.

If acceleration cannot be made persuasive to people there is a reason why and if you cannot make it persuasive that is a signal of the fitness of your coalition. In conclusion: true e/acc is likely to be pro-democratic even if it must trade some short term gains
0
0
8
Great video of @willcb on a real outlook on how continual learning can actually be achieved on a meaningful scale, backed up by companies that are actively doing it

youtube.com/watch?v=5SRjPT…
0
0
4
Yeah really want to see some Hyper-Abundance Techno-Progressive Optimist Libs now. Like, lean into the aesthetic, but also make it happen.

We have to start the process of realizing some form of UBI now if we're going to have an honest conversation about automation and AI in the slowdown that some dems are signaling.

We have to start the process of building a national retraining program to take folks from jobs that are in the process of being automated away, as well as ones that have been or will be, and finding them meaningful, gainful employment elsewhere in the economy. Trade schools are a great and popular option across the aisle.

We have to start the process of protecting the future democracy from the same things happening all over again, both rampant expansion of the surveillance and police state and illegal and unpopular wars. We can't wait it out and hope the American public won't vote for it again, because they absolutely will.

What we need now is for the purveyors of the peril we have faced to pay, for real protections that have teeth to be put in place to ensure we don't face these exact same problems in 2, 4 or 6 years and for a real, optimistic view of our future, not just another referendum on the incumbent party.
I was just thinking on my drive today that R imagery going into 2028 should be futuristic renderings of hyper-Miami, etc.

Romanticizing decaying diners in abandoned small towns and talking about an "America of 100 million" is out.
5
4
26
A reminder: acceleration is a vector.

vectors have magnitude and direction.

e/acc is accelerationism that treats itself like a scalar in public, only talking about its magnitude as if the scale of acceleration is all that matters.

pay attention to the vector direction.
how are backwardly looking conservatives suppose to be the ones to accelerate us their bullshit makes zero sense
0
4
21
Remember, folks, this is what @varien built groktheblock.com for: so you don't have to waste time making multiple clicks to find out someone's opinion on US politics should be routinely and aggressively ignored
1
1
12
the blue wave started with the Knicks winning the NBA championship.

in this essay i will...
0
0
5
Southerners will be like

"Yeah New York is alright but they can't beat the south on food"

and they're just talking about Greasy B's
0
0
3
Currently, @scott_colom , a black Democrat, is leading Cindy Hyde-Smith, the incumbent white Republican and MAGA darling, by 42 votes in Mississippi.

This is what the county-level votes look like.

When MISSISSIPPI is swinging this hard, you should listen to the message.
0
0
3
Bleu Wave + Smaltman cope + Community Note

Who needs coffee when you have this
i want you to have so much ram

lets build more
0
0
4
BREAKING: Anthropic has updated their S1 filing once again after seeing the results of the midterms.
2
1
36
Alec Baldwin's nightmare looks pretty nice
Alec Baldwin says it’s been a “nightmare” living under the Trump Administration.

“We’ve been living in this nightmare, for me and my political consciousness- it’s been a nightmare…The beginning of the end of that nightmare is tomorrow.”
0
0
7
Slowly descending through the clouds, back home away from civilization
0
0
4
Bullish on @NousResearch

Sad to be leaving NYC. Happy that I will definitely be back soon. Was happy to get to show the kiddos the big city too, they're more sad to leave than I am
4
0
103
This is a NY fan account for 1 more day sorry I dont make the rules
2
0
14
Apparently in NYC you can get on the subway (very cheap) for 20 minutes and pop out in what feels like a totally different city (east village)
7
3
138
Mississippian in New York talking to other New Yorkers about Mississippi
I have been in NYC for less than 24 hours and I immediately understand why red state Republicans need to fool their constituents into thinking this place is a hellscape dystopia
1
1
19
Like come on you are gonna try to convince me that this place sucks
9
2
77
Damn these mfs really dont sleep
0
0
20
Legendary night at NousCon. Was awesome meeting @karan4d and @Tom_A_Lynch and finally hanging with the homie @kojinglick and so many others
1
0
43
Holy shit NousCon is dope
7
6
111
Replying to @valhalla_dev
I have eaten bodega sandwich.

It was heavenly.

This city is amazing
0
1
10
Replying to @valhalla_dev
Damn this place is big
2
0
3
Mississippi Boy visits Real Civilization for the first time in a while/NYC for the first time thread

First impression: Public transport is neat
4
0
28
Hell yeah this from @WillManidis vibes with what I wrote about this week:

The firms of tomorrow will be building systems instead of products.

x.com/valhalla_dev/s…
An optimistic theory on systems development replacing product development and a manifesto/hypothesis on cybernetics

or maybe just cope from someone who is ngmi, who knows x.com/i/article/2113…
1
1
3
T-25h NousCon Hype Train
1
0
2
tonight, we learn about hill climbing.

youtube.com/watch?v=-iu2-s…
0
0
0
I feel like this type of video is so unbelievably rare in the sloposcape that it deserves constant calling out

h/t to @bennjordan for this work

youtube.com/watch?v=OP5GDi…
0
0
1
"Hey Developing, not gonna waste your time."
1
0
0
The normies are talking about bare metal, de-servicing and sovereignty. The culture is shifting.

ft. @SimonHoiberg
youtube.com/watch?v=cdxQRF…
0
0
3
huh, slightly less dumb than I thought I'd get which is reassuring.

thought for sure that my distractability and garbage memory would kill me on this one
huh, this matches up almost exactly with the professionally-administered IQ test i was given in high school
0
0
0
Me explaining that Elon, Trump, Bezos and Michael Dell are actually more Communist than that crusty barista with 4.5M Reddit Karma who did his own hammer-and-sickle neck tattoo
weird as it is, the richest people on the planet discussing giving away shares of a company directly to people is the biggest advancement towards collective ownership done in at least a few generations
1
0
9
tough mudder 9 miler in Nashville in the books with wife, lil bro and friend.

only electrocuted 3 times, 10/10 would recommend
0
0
8
Why are we still falling for obvious funnel bait
0
0
2
~a few hours until The Homies drop some insane tech
2 days until 9 mile Tough Mudder race in Nashville
5 days until @edgerunners Season 2
8 days until @brakence album drop
8 days until @polyphia album drop
15 days until @NousResearch NousCon in NYC
0
0
6
2021 btw, before we sent most of the US Navy to Hormuz and bombed the hell out of the country for no reason
Yalls continued use of AI (chatGPT, Grok, Tako, etc) is the reason we have 100° weather in October btw.

I'm being so serious, y'all are fucking up the planet and you don't give a shit because you can't use your brains for simple tasks.
1
0
0
`Lint is currently misleading. The API and web lint scripts literally run echo 'lint: ok'. Green “lint” jobs are not source linting, so the existing gate does not meet the intent of your new SOP.`

idk which model did this but it is both funny and getting tossed in the model torture chamber
0
0
0
Who has the highest lifetime token usage for Claude Code or Codex?

I'm finding that apparently I do not eat tokens like most of the people I know.
3
0
2
Guess I'm going to be in NY at the end of the month
1
0
10
Replying to @NousResearch
c o p p e d
0
0
5
"agents want to <xyz>..." - silly; relies entirely on people believing that matrix multiplication is conscious enough to have wants.

"we're studying emergent properties and behaviors in multi-agent social systems" - honest, interesting, not wooky
0
0
1
Super good practical breakdown on observability

youtube.com/watch?v=AmHYfE…
0
0
2
running with bronchitis kinda sucks
0
0
0
things seem to be going well
0
0
4
They nerfed the Pro 200 tier just to drop a new Pro 500 tier, and they released "OpenClaw but with even lamer aesthetics"

Massive OAI L on DevDay
OpenAI have accidentally hit publish on their complete DevDay summary early

openai.com/index/devday-2…
3
0
36