Don't stocks fall after all time high basically by definition? www.cnbc.com/2025/09/21/s...

Moondream 3 A 9B param, 2B active MoE vision hybrid reasoning vision language model that supports both reasoning and non-reasoning mode. It focus on visually grounded reasoning, where the model references objects and spatial positions in the image while doing said reasoning,

The galaxy brain is that military genius is just the abilty to train and move more reliably average soldiers than the other guy can

producing successes is great. really important, success. you know what's even more important? surviving failure. because no matter how good you are, eventually failure comes knocking. and if you can't survive it, that's that.

We are addicted to grievance

Is there a "total" war where this hasn't been true? Us civil war it certainly was, ww1/2... but I'm no military scholar

I do find it funny how so much male gaze pop culture history about Rome is obsessed with painting them as inspired military geniuses when the real lesson of Roman expansion, including in the Punic Wars, is that military genius ultimately fails against an infinite supply of reliably average soldiers.

Posted in support of h1bs and went viral and only got one hateful dm. Maybe there's hope for the world (I'm not going to check the replies)

i was privileged to become american thanks to the h-1b visa. the program has its flaws, but adding a $100k "bribe" (which can be arbitrarily waived if you please the right people) is not the answer

On the face of it this seems kind of insane. But then again, what exactly do I pay for that isn't one of these big techs product or service? Basically just food and housing

Chris Freiman talked on Twitter about how downplaying the economic strength on both the left and right is used to boost support for radical reforms that aren't necessary or harmful. I see a lot of this here.

whoah, TIL

I am an active urban cyclist and I feel nothing but completely safe around a Waymo. My head is on a swivel if it's a human, especially if they are in a pickup truck or SUV. Waymo? They'll stop on a dime or safely route around you to avoid a comical amount of dangers.

Self-driving cars should start being the default. Humans are complete dogshit at driving -- we could eliminate so much property damage, injury, and death by providing a streamlined regulatory environment to get more companies like Waymo up and running. www.understandingai.org/p/very-few-...

K-Dense: A multi agent that has made discoveries in human aging - Gemini-2.5-pro - 4 specialized sub-agents - discovery improved aging by 4.5 years - needs guidance by human researchers www.biorxiv.org/content/10.1...

This is far from a rare viewpoint; this is probably the best way to get a massive dataset for embodied ai training.

The US president saying "Comedians and the news are not allowed to criticized me" should be giant front-page news in every outlet in America, by the way. It's him repudiating the First Amendment and the Constitution. Unambiguously and clearly.

This is the top headline on CNN. A good warning sign that your country may be becoming a dictatorship is when your plain-jane straight news networks start saying “Our country is becoming a dictatorship”

Figure just raised over a billion In this blog post i estimated that building robot GPT would be a ~$1 billion USD project for a major company if you wanted to do it all in the real world so i guess here we go open.substack.com/pub/itcanthi...

It is obscene that none of the people behind this are in jail, and in fact are back in the whitehouse. www.theatlantic.com/magazine/arc...

If grades and classes don't matter, students need other ways to stand out which makes life harder and more stressful -- all the while denying everyone a good education. www.theatlantic.com/ideas/archiv...

Seems bad

You can always turn up temp to get your base models to produce anything right? But the core issue remains that they need to assign high enough probability to good stuff and low enough to bad stuff that exploration can be done in meaningful time

This I think is something lots of people don't get about RL

If your base model is unable to solve a task, can you simply apply reinforcement learning to fix it? The answer is no. For reinforcement learning to work effectively with language models, the base model must already assign a non-negligible probability to producing a correct solution.

First California then the world

Imagine being this guy. What a blessed life he must live

The advantage is also pretty minimal compared to other fields

Above all things continue to get better

good

infuriatingly wrong from our (former) CPO

The 90s techno optimists foresaw -- correctly -- that the internet would be a font of knowledge and would bring people with common interests together. They missed that people would become addicted to the same kind of cheap on demand misery that had already started piling up on cable news.

No doomers, no crab buckets, no techno-pessimists in this tent. The future will be better and weirder.

I dont, to be honest, want to care about politics. I do care, quite a lot; I'll be a liberal to the last. But more than anything I want to live in a good world and leave a better one. Political violence unravels all of that, and every assassination I fear takes us one step closer to the precipice.

People who shut their eyes to reality simply invite their own destruction, and anyone who insists on remaining in a state of innocence long after that innocence is dead turns himself into a monster. — James Baldwin

Very exciting

Humans in their role as summarizing and cherry picking machines

US sources for these products don't exist, aren't of quality, or are cost-prohibitive. S-100 PCBs for a CompuPro: China Power supply, VGA, & ROM upgrades for an Atari ST: UK Vectrex upgrades: Australia & UK Console flash carts for dev: Ukraine Oscilloscopes: China Soldering gear: Japan

"never use bots as therapists, you absolutely need a full blooded human for these kinds of problems!" the humans you're paying $250/hr for:

Might as well just pay for a ChatGPT subscription at that point www.technologyreview.com/2025/09/02/1...

Baidu, China's Google, released ERNIE X1.1 Per them, "✅ Achieves significant performance gains over ERNIE X1, with factual accuracy up 34.8%, instruction following up 12.5%, and agentic capabilities up 9.6% ✅ It surpasses DeepSeek R1-0528 and performs on par with GPT-5 and Gemini 2.5 Pro"

Any deficit in blog posting can be blamed on Silksong coming out last week

A German teen told Meta's child safety researchers his under-10 little brother had been sexually propositioned multiple times on its VR platform. Meta deleted the evidence. New internal whistleblower docs indicate that was part of a broader cover-up: www.washingtonpost.com/investigatio...

This is good. wild that they can easily afford it though.

Bluesky is shrinking rather than growing. It has no source of revenue. If it doesn’t grow, *it will die*. Constantly making it unwelcoming, saying who you don’t want here, setting rules for what other sites people should and shouldn’t use, and the like, will kill it quicker.

Exa raised $85M on a $700M valuation this is one of those things that just makes me happy. they’re such a good company exa.ai/blog/announc...

i have heard so much whining about golden handcuffs in my life

i do not understand the mindset required to make $300k and spend all of it! it's just baffling that people who have never experienced poverty nonetheless put themselves in a position where they are in perpetual fear of it.

You don't have to theorize about why Mark Andreessen gives money to abundance people. He wrote a whole essay about it 5 years ago. a16z.com/its-time-to-...