A\ is sufficiently different from previous envs. my previous methods aren't as effective as i am used to think, so i feel like i need to up my neural plasticity game, and need to reset myself. it feels like a steeper-than-usual hill to climb, but worthy of the challenge
i've changed a lot in the past 15 years. it was a mutli-step process: each new env (US, Google, Amazon) setup a new hill / reward metric for me to climb. In hindsight they were pretty similar still, in the common direction of software engineering.
My 8yo says he chose us, parents, before his birth, because are worthy
The zero abstraction principle means the stuff can be fast and you won't have to rewrite it later. Without this, you think "I have this great idea, and I want to implement it now but if I do it in Python, then I'll have to rewrite it later anyway so why try hard if i will throw away anyway."
I'm so glad Niko Matsakis visited our team and convinced us to choose Rust over Go back in Amazon. This led to me learning Rust which then taught me things I never thought about, such as ownership and lifetimes. This now influences how I think about programming.
Why wouldn't they be able to talk to each other? Do you mean people are able to tell whether they are talking to a real person or not?
Yeah, we've fixed it a week or so ago
Selection bias, unrepresentative model?
You can also imagine a Rust macro where you write a prompt and it computes whatever you want
> I'm looking for a new job, with a preference for larger companies that require in office attendance. ooc, why?
I thought the worst part is to die each time
It gets better!
It's just so happens that the names of 8yo and 18yo start with "Ai".
Oh wait, you said Advanced Voice, not just Advanced. I think my reply can be ignored.
Best of N I mean. Run inference multiple times, then pick the best result. This would explain how you can enable/disable advanced mode in the same chat without changing the model.
Maybe just BoN?
I liked Silo S1 on Apple TV. Recommending
I don't like the present *state*, but *present* might be the only real thing, i.e. the past/future might be a probability distributions consistent with the present
What exists? What's real? Seems like any state from any continuous realm can be approximated in a discrete realm and thus can be mapped to a binary string
Everyone needs a Rust therapy once in a while
Falling in love with Florence + the Machine
A common misconception of Staff engineers wanting to join @anthropic.com is an expectation that they will continue doing the Staff kind of work that they did at a prev job A half of engineers here are former Staff/CTOs/Directors at other companies. The right expectation is to roll up their sleeves.
He tanks the stock market, and he and his rich buddies buy low… Then he pauses the tariffs, and stocks rise… He and his rich buddies sell high. Rinse and Repeat…over, and over again. It isn’t Rocket Science… It’s actually criminal.
Everyone is on MCP now
If I were applying to college, this would be my application
We have like 1-2 years left. Gotta code while we can
To be fair, that's Llama 8b, a tiny stupid model
What about search inside chat apps?
Ugh I get it now. The true villain of the plot is the AI companies. Shit
biggest problem with python is that it enables others to write insane code and then i have to deal with it
Trump impact on me: the ETA of a green card petition I filed increased from "12mo" 9mo ago to "20mo" now
Can we understand the mechanisms of a frontier AI model? 📝 Blog post: www.anthropic.com/research/tra... 🧪 "Biology" paper: transformer-circuits.pub/2025/attribu... ⚙️ Methods paper: transformer-circuits.pub/2025/attribu... Featuring basic multi-step reasoning, planning, introspection and more!
The game "Split Fiction" is just great. So well made and fun to play with my kid.
I'm so glad my aunt hasn't sold the house I grew up in. Of course I had take the old route: subway, a walk in this old neighborhood. Nostalgia full on
Visited the classroom where I realized that I'm really grokking programming 25y ago. The principal of my age was accommodating. The subway stations didn't change. Train cars did. Coins were replaced with QR codes.
Went through major nostalgia by visiting my school where I spent 10y and the subway station next to it The voice announcing the next station hasn't changed for at least 30y!
Barbwire around American University of Technology, affiliated with ASU 🤨
Bizzare to hear the music of my childhood in cafes. It is not some national music, but American music from 90s/00s
Today was eventful. During the passport renewal process, employees of Uzb's FBI equivalent got extremely interested in my Google/AWS background and specifically AI expertise. This was a long conversation.
The word "IT" has a different meaning here and it confuses me. It encompasses all tech, including software engineering and now AI. I'm considered an IT guy Lip fillers seem to be more popular than in US. I've never seen lip fillers in Uzbekistan before.
I just remember an LLM complaining about the quality of the code. I think if you encourage that in a system prompt then it will do it
I never spoke Uzbek, and I do sometimes speak Russian at home. But what I had to recall is how to write in Russian on paper with a pen. I think I didn't have to write it for maybe 10y. For example, I accidentally wrote u instead of Russian y in my last name. The two letters have the same sound.
Try to prompt it about this. It can be fun
Spent half day in a government facility trying to renew my passport. Exhausting.
People smoke in public places, like in the airport restroom Taxi prices: $1-2
Arrived to Uzbekistan. Haven't been here for 5y. It's different. So many Chinese cars and phones
Anthropic employment is a very humbling experience
📌
happy 3.7 day to those who celebrate
Android lets you combine two emoji. For some reason.
If you are right then this gives some hope for us, humans, and our purpose. Either way, a win.
Oh what did Larry do?
> you can enjoy your vacation when we can enjoy our democracy Brilliant, no notes
PSA: use CLAUDE.md, in the project dir or homedir, to instruct Claude Code how to behave, code style, desired way to engage with you, anything. docs.anthropic.com/en/docs/agen...
Monday was fun: - Launched 3.7 Sonnet - Wife started at AWS, first job in 10y, since moving to US - Son's birthday
Depends on how good Gemini is. Public transportation is cheaper than a BMW too
Your annual reminder that because February is 9% shorter than January was, your monthly metrics will be down nearly 10% this month, and not to freak out about this.
hespera has found God eos has found Claude
Nature has developed agents for us
I take it a step further: ask Claude to make an artifact with interactive checkboxes. Works on the phone. Converted an iMessage screenshot into a single-use app
Claude Sonnet 3.7 with Claude Code has been a great help pairing with me on MCP SDK work. So stoked to see what people will build with it. www.anthropic.com/news/claude-...
Given how much engineering is required in modern large-scale AI, the impact much be less than it appears
www.bbc.com/news/article...
His argument is strange, as if AI is a single/few entities. It's like saying Elon buys the Internet. Too large to be contained, too attractive to have a monopoly.
> you're letting the keepers of that AI have power over what you learn and how you write How is this different from school system? Don't they have the power of what kids learn?
Logging into X brings sadness
Such a typical human behavior to design everything after themselves: gods, robots. A car is a more practical specialized form of a robot. A house is another form.
Not even six hours apart.
Code simplification is probably my favorite activity in programming
🎉We are pleased to announce that the Java MCP SDK developed by Spring AI is becoming the official MCP Java SDK!🎉 spring.io/blog/2025/02... As the MCP core team, we are happy to have an SDK for Java that feels like Java and fits perfectly into the wider Java ecosystem such as Spring AI.
Need slack reacjis in bluesky
Seems relevant www.anthropic.com/research/mea...
ooc what's the context?
T stands for the
Crypto guy in 2014: Crypto will revolutionize finance and change how we interact with money Crypto in 2025:
I'm honestly not sure if I ever worked so much, but it is fun and rewarding
How intelligent do you expect such a model to be? I haven't worked with so small models but I imagine they can't do much more than producing intelligently-looking nonsense?
why do people join openai
Finally sat down to build a swarm of agents to deal with my personal stuff
Phew this is a spicy one: DeepSeek apparently had a misconfigured DB that allowed anyone to connect to the DB and query chat conversations. Huge privacy breach! And that's why you run models local, or use a trustworthy API like Bedrock, instead of using startup APIs: www.wiz.io/blog/wiz-res...
I thought this is about the simulation theory before I read the text
I just want to live in a world where actual Nazis are imprisoned, not pardoned. Brave new world.
Prob this year
if we were in a simulation indeed, would the operator understand what's happening in our minds, or would they be like mechanistical interp folks?
"Siri, where are my pants" is still not working
MCP is the future AppStore for LLMs
I am looking for passionate software engineers who have experience in maintaining Open Source projects, and want to work on Model Context Protocol for a few months. You would have a strong track record of open source contribution, care about AI safety and know TS and/or Python
I think most AI people realize how hard it is to predict this stuff, so predictions are usually taken with a default grain of salt. I didn't find anything controversial in these predictions - they seem safe
It's so nice to be able to send a slack msg at 11pm without thinking "oh is too late? Should i schedule it?" No it's not b/c ppl is working at night. It's cause the receiver trusts that I am not pushing them to work late and know that I won't have second thoughts about them not responding
This topic is weirdly fascinating to me. For decades I thought computers can't do anything beyond following simple logical rules, probably influenced by my father, who considers computers inferior due to their discreteness. AI demonstrated that they can do a lot more than I thought.
Apparently not everyone thinks this way and some people are quite open to the idea the world is discrete at the most fundamental level, e.g. at Planck constant level. Ultimately, even if the reality is fundamentally continuous, it seems that it approximated well in the discrete realm.
One assumption I have deeply ingrained in me by my classical physicist father is that the reality/world is fundamentally continuous (not discrete). I'm not a physicist, so I don't have my own opinions/experience on this and never questioned it.
Ok now you have to show us the video that you recoded that was less cute than in real life
