Dario Amodei on open source models: “If the path continues, I think we could get to a very dangerous place.” “When you control a model and you're deploying it, you have the ability to monitor its usage.” “You can revoke a user's access. You can change what the model is willing to do.” “When a model is released in an uncontrolled manner, there's no ability to do that. It's entirely out of your hands.” “We should confront the advocates of this with that problem and challenge them to solve it.” He clearly doesn't believe Intelligence should be in the hand of the masses and Sam Altman also saying "RAM" shouldn't be accessible to everyone indefinitely agreed with him.

Steve ☕️ ✓
@coffeedevdev, fixer, tinkerer


"Everyone should run their own local AI" is the new version of the "everyone should self-custody" delusion. Local AI is great for niche power users. Get out of the bubble.
The vast majority of people have zero interest or desire to run, maintain, manage their own hardware and services. The local ai argument is largely the same as the crypto self-custody argument. It's great in theory but a lot of time, money, effort to get it right. The vast majority of people are going to use hosted services and be happy with it. They don't care about self-sovereignty, uncensored models or privacy.

Okay, what does it mean that ATC "lost frequencies?" Radio frequencies don't just go away. This is a 100-year-old technology. I don't understand the term the way they're using it.

Can someone explain to me how a "frequency" goes down? Radio waves are physics, they don't just stop working.


Before claude I was oblivious to what an "eyebrow" was in terms of design. Now, not only do I know what it is but I despise it with every fiber of my being. Is there a name for this phenomenon?

*preach decentralization and CROPS* "Just make it upgradeable for when we break the chain and your stuff doesn't work anymore" ???
Never ever break an existing contract is not workable. It would mean you cannot change the gas schedule. You also cannot deprecate signature schemes which we will most likely bave to do in the future. Instead, we should be clear what kinds of contracts will not break. If you rely on gas schedule or state commitment, for example, you can't expect your contract to work forever and it needs to be upgradable or its users need to be aware that they have to exit under some circumstances.


Who is building the decentralized world computer with support for immutable smart contracts

"Make sure your contracts still work after the upgrade" and if they don't? is the protocol going to give me a way to alter their bytecode to make sure they continue working? absolutely batshit crazy work by core devs
PSA: Please verify that your contracts are still working after Glamsterdam! We created a repository with Invariants that will be broken after the fork, so you can verify for yourself. blog.ethereum.org/2026/11/06…

can someone give me the tl;dr on why everyone hates protonmail all of a sudden seems like I've seen a ton of negative sentiment all of a sudden out of nowhere

if a video is obvious AI slop, but I still get a kick out of watching it, is it still "slop"?

definitely couldn't be me
>be me >be 34, two kids, one wife, one nas >buy second 3090 off some guy in a walmart parking lot >he counts the cash twice and says "god bless" >finally run qwen 4 27b fully local >q4, 190 tok/s, room is 31 degrees >no cloud, no api key, no telemetry >"nobody can read my prompts now" >first prompt: "is the government about to collapse" >model says "that depends on several factors" >mfw $1400 of silicon to get a weather report >midterms tomorrow >every ad is a man in a flannel standing in a field >every other ad is a woman in a flannel standing in a different field >can't tell who's winning, can tell flannel is winning >meanwhile some agency puts out a "preparedness advisory" >"citizens should maintain 72 hours of food and water" >they literally say this every single year >but this year the news says it in the SCARY font >wife: "we already have 72 hours of food" >me: "that's for the KIDS" >drive to costco >everyone at costco is me >eleven guys buying the same pallet of water and the same 50lb rice >we make eye contact >we do not speak >we all know >get home, ask qwen how long 50lb of rice lasts a family of four >it writes a python script >the script has a bug >fix the bug myself >mfw I am the inference now >3am, can't sleep >new neighbor moved in last month >keeps weird hours, foreign plates on the car, carries boxes in at night >"anon.exe has detected a sleeper cell" >ask qwen if my neighbor is a sleeper cell >"I can't make that determination from the information provided" >based and lawyered >stay up anyway >watch him through the blinds like a tactical hamster >he's carrying boxes in >boxes say ZOTAC >no >NO >he has 3090s too >he's doing local inference too >he's not a sleeper cell he's a SLEEPER BUILD >next morning he knocks on my door >"hey man, your rig trips the breaker at night, we're on the same panel" >"...how do you know that" >"because mine does too" >he's an electrical engineer from lahore, moved here for a job the company couldn't fill locally >he has a 4x3090 nvlink setup and a 5kw ups in his garage >he also went to costco >he also bought the rice >we stand in silence >we both know >he runs a quant I've never heard of >he benchmarks ~360 tok/s >mfw the "sleeper cell" is just a guy who out-built me >mfw the immigration crisis is that my neighbor has better airflow >midterms tomorrow >polls open at 7 >we're going together >after we vote we're going to pool our cards and run the 235b >the war can wait >the rice can't >tfw the real national security threat was a 200 amp panel from 1974 >tfw I still don't know who's winning >tfw flannel

RIP to everyone that needs a microtik switch for their sparks Guess we are running rings for the foreseeable future

Amazing news! KindlingAI is joining the @NVIDIAAI Inception program. We've been working non-stop delivering the low-KL, high-performance EXLR8 quant format plus the gr8 engine, and we haven't taken a breath. 1.1M downloads of our GLM 5.3 weights on huggingface! Almost half a million downloads of our GLM 5.3 Flash weights too. We're continuing to add new models, new kernels and new quantization techniques to the KindlingAI library, and we've been supported by a bunch of awesome community members too. Thanks to @abheer, Alex Steiner and the other NVIDIA folks for all the support along the way. We can't wait for you all to see what we're doing in the next we weeks and months. ❤️ Join @CK2084, @coffeedev and me on our adventures!










