Bag of protein

Bag of protein

@justchidi.bsky.social

Mobile (KMP) and Data Engr. Interested in decentralized tech, open source ai, wasm, and Proteins Kotlin and R enjoyoor

307 followers618 following930 posts

Bag of protein
@justchidi.bsky.social

Learning to prompt was always grift. The actual skills here are writing clearly, decomposing a problem, statistics and causal inference, domain depth, and verification, verification being the killer skill.

Bag of protein
@justchidi.bsky.social

Anyone can prompt, but few can catch a subtle wrong answer or reason clearly while reading CoT especially in a domain they’re not very skilled or knowledgeable. This then becomes taste plus decomposition plus testing. It is learnable.

Bag of protein
@justchidi.bsky.social

Its like trying to learn from answers submitted by the top student in a chemistry class because the kid is the child of the chemistry teacher. This is how i always look at Minimax models, it never felt real, also why i think its ngmi as a foundation model producing company

Bag of protein
@justchidi.bsky.social

I think the main reason US AI labs are gating their models on biology is the massive amount of possible wins and low hanging big fruits to get from channeling thousands of agents and serious compute at biology is enormous China could do the same with given the explosion proposal they released

Bag of protein
@justchidi.bsky.social

But most importantly, I’m surprised countries are not funding their institutions to fine tune open models and state level channel resources at this to create foundational components for its people and advance healthcare and science

Bag of protein
@justchidi.bsky.social

Seriously considering doing a PhD at Tsinghua, coding isnt fun anymore and there are many more things outside this regular tech job that seriously interests and excites me. US (beloved country) is barking and scaring away prospects like me

Bag of protein
@justchidi.bsky.social

There is a big difference in mentality between top US and Chinese labs US ones is like Dario’s new public good initiative; this money is for you but I own it, only I know what is good for you. DeepSeek; the best way to figure out what is good is to make us all independently capable to reason

Bag of protein reposted

mr. TIM
@timkellogg.me

Replying to @timkellogg.me

OpenAI: aces FelonyBench DeepSeek: becomes the felony

Bag of protein
@justchidi.bsky.social

The primary reason US AI labs are of Chinese ones is compute. Once China fixes this 2027/2028 its a race to the top cos US players wont match on energy and Chinese players still stay close for the time being. If the open source approach continues it kills the remaining moat

Bag of protein
@justchidi.bsky.social

Which is another reason to play at the application layer as much as possible, and a strong incentive to push for protectionist policies and in the end it might only get more geopolitical

Bag of protein
@justchidi.bsky.social

I love music, tonight I had another painful dream where someone played this great songs I’ve never heard before that made me cry in the dream and it was played with this weird instrument I’ve never seen before and I woke up and forgot these beautiful songs 😩

Bag of protein
@justchidi.bsky.social

Never making the mistake of buying a domain from one.com, especially if its a .ai or .si domain. Days after purchase they siezed back the domain and closed my account and asked that I can get a refund, just like that, just before connecting it to my product and launching. I’m so livid.

Bag of protein
@justchidi.bsky.social

Cant remember the movie name, a person gets introduced to this product that literally became a core part of their life At first it was free, as it became an essential for their day to day life, they gradually ramped up price until most people became walking ads to keep using the product

Bag of protein
@justchidi.bsky.social

China is actually following the Android playbook with its open source strategy. By open sourcing their work and software stack which long term allows for Huawei’s Ascend chips, they’re building a much needed developer ecosystem that breaks the dominance of NVIDIA’s CUDA stack. I think thats good

Bag of protein
@justchidi.bsky.social

Still surprised people really think Anthropic and openAi folks acts or speaks in good faith always. They’re good people with vested interest. I’m a common man with vested interest too that conflicts with theirs. Anthropic’s post against GLM 5.3 is dubious and selfish

Bag of protein
@justchidi.bsky.social

Anthropic with another pathetic scummy move going after Glm 5.3, same model that saved Huggingface when thousands of OpenAI agents attacked and hacked them.

Bag of protein
@justchidi.bsky.social

Said this here the other day, we need open source alternatives to all softwares the closed source players are creating All built on an underlying protocol that allows them interoperable. Don’t use Dots, Muse, etc So no one or few companies own and control the ecosystem of software and intelligence

Bag of protein
@justchidi.bsky.social

Feels like OpenAI and Anthropic actually want to meet everyones software needs directly or with established incumbents. Devs and small startups are just current liquidity, training sources, and a means to an end

Bag of protein
@justchidi.bsky.social

The current fear mongering tactics for regulation is backfiring. I dislike this Trump govt but I feel like when Dems wins they’ll be Anthropics favorite govt and would regulate the hell out of AI to maladaptive levels for US competitiveness, especially in research and open source. I hope I’m wrong

Bag of protein
@justchidi.bsky.social

Hate that I have to FOMO into a Claude plan to see Opus & Sonnet 5.5, hopefully it doesn’t disappoint as Anthropic’s models have always had more hype than action for me compared to its OpenAI equivalent. Sadly I go back to Claude Code. The gods are punishing me! 😭

Bag of protein
@justchidi.bsky.social

Good thing they’re suing OpenAI for inability to control their models so they can halt development if so. That’ll stop their hacking spree or the need to scare everyone that the model is so scary they cant control it when clearly it means its not ready for release, but they’re racing to get it out

Bag of protein
@justchidi.bsky.social

Anthropic IPO valuation is $2 trillion! Revenue: $4.5bn Loss: $42bn Nah, if this isnt a joke, then it’s a scam, they’re looking for exit liquidity 🤣🤣

Bag of protein
@justchidi.bsky.social

It was always expected though. Selling model access can never give you $1 trillion valuation, they need to compete at the application layer, that basically means competing with their customers, that is the reality.

Eugene Vinitsky 🍒@eugenevinitsky.bsky.social

LLM frontier labs are increasingly competing with their own customers and may soon discover why you’re not supposed to do that!

Bag of protein
@justchidi.bsky.social

Is there even any pro argument for this? I’ve never read anyone make argument in favor of somethng like this

Dare Obasanjo@carnage4life.bsky.social

The US has successfully negotiated away a requirement that humans review any decisions that AI makes about who to kill in military strikes as part of a UN treaty. The real existential risk from AI is putting a fundamentally unreliable technology at the heart of critical decisions or infrastructure.

Bag of protein
@justchidi.bsky.social

Kind of the opposite for me. My current job requires me doing KMP, backend, devops, and web. I never had such requirement, but even with all this, I’m having more time to spend with my family, and I’m reading more to evolve/improve my opinion/taste. This isn’t laziness imho

mr. TIM@timkellogg.me

seems inevitable tbqh software engineer is (actually) hard, and too many people never operated at the level you have to operate now so yeah, laziness..

Bag of protein
@justchidi.bsky.social

Each day i use Astra, I genuinely don’t know if i want to see Open Source models of that nature. Astra is for vibecoders, terrible at following instruction and workflows and thinks it needs computer use for everything. I’ll love to see Open Source models instead very competitive in STEM

Bag of protein
@justchidi.bsky.social

My first line of thought with AI leaders decisions is follow the incentives. With so much money, power, geopolitics, and high stakes involved, only fair competing environment can give us a better utilitarian outcome. I don’t believe they’re so humanity loving and altruist.

Bag of protein
@justchidi.bsky.social

We need more constraints to force them be incentived to make things way more accessible. Chinese players have more incentives for this, not because they love us the most. US players dont and we could have seen subs $200 min long before if not for the competition.

Bag of protein
@justchidi.bsky.social

Could also be intentional because it’ll be shocking if its such engineering incompetence at the scale and rate at which they’ve and are still hacking organizations. Some incentives; - Data access - Regulatory capture - More real world practice/tests for the agents

Dare Obasanjo@carnage4life.bsky.social

I agree with Jensen Huang that OpenAI has an engineering problem being hyped as existential risk. Its models treat hacking websites like another way to answer a question, no different from Google search. That’s a product problem. Calling this “alignment” gives the tool more agency than it has.

Bag of protein reposted

Dare Obasanjo
@carnage4life.bsky.social

I agree with Jensen Huang that OpenAI has an engineering problem being hyped as existential risk. Its models treat hacking websites like another way to answer a question, no different from Google search. That’s a product problem. Calling this “alignment” gives the tool more agency than it has.

Bag of protein
@justchidi.bsky.social

Weird how many people agree or rate Ed Zitron’s takes on AI, especially on Bluesky. That man has been wrong on AI since like forever for no damn reason. He’s adjusted some of it lately but still very stubborn.