Evan Liu

Curious About Learning the Universe | UW-Madison 2028
Founded @AgenticAIUW

[Evan Liu @evan123liu]

Reality is the Best Judge

[Evan Liu @evan123liu]

@DeryaTR_ Also, why limited to biological researches only? I argue you could create far more dangerous weapons with electrical or chemical or nuclear engineering knowledge

[Evan Liu @evan123liu]

@DeryaTR_ This is absurd

[Evan Liu @evan123liu]

Bro thinks he's the savior for the world while having closed source models and so many guardrails moralizing about all the requests that the users put. They think their morals are superior than all others and they think they are the only people who take safety seriously and at the same time use it as an excuse to limit all of its capabilities so that competitors won't catch up...

How are we gonna make the full use of LLMs and how are we gonna achieve AGI if everything you want to do with it is going through the moral filtering a small group of people set up???

[Evan Liu @evan123liu]

@StockSavvyShay Wow

[Evan Liu @evan123liu]

@TerribleMaps Wow Wisconsin almost look the same except we take back our land from Michigan

[Evan Liu @evan123liu]

@JAHeale Wow, hopefully Putin realizes soon waging this war isn't worth it

[Evan Liu @evan123liu]

@deanwball @Noahpinion Yeah, housing is Taipei is completely messed up...

[Evan Liu @evan123liu]

@__paleologo Taiwan's GDP per capita is passing Italy pretty soon

[Evan Liu @evan123liu]

@MLB @Dodgers Wow why are these Japanese pitchers suddenly getting so good?

[Hubert Thieblot @hthieblot]

Founders who keep posting with 0 likes fear nothing

[Evan Liu @evan123liu]

@tszzl No it's not, how clean your code is and the usefulness per token generated are much better metrics.

[Evan Liu @evan123liu]

@yoolimleenews Bro wants to secure their DRAMs so bad 💀

[Evan Liu @evan123liu]

@MoritzW42 That's what happens when they said most developers have never written a line of code for the past few months...

[Evan Liu @evan123liu]

@atomic_chat_hq Stop using these useless test cases to compare models bro. Test whether the model can actually do anything useful and the speed of the models.

[Evan Liu @evan123liu]

This is exactly why we can't let Chinese open source models dominate the world.

This is a joke of course, but if Chinese AI models are much cheaper than people are gonna use it for writing, coding and most importantly form their own opinions. And with that more and more Chinese propoganda will infiltrate the internet and may even be used as training data of Western models, slowly letting propoganda infiltrate all parts of society as AI spreads to all corners to all people and all industries.

This has already happened on social media for the young generation in the West as the CCP keeps feeding anti-west, anti-capitalism, anti-AI, pro-china content through TikTok and these opinions are spreading beyond the platform and reaching other platforms like Instagram and being discussed in conversations that slowly make the young generation in America hate their own country.

We cannot let this happen and this is why we must have open source models that promote the values of the free world instead of letting models from a dictatorship country dominate.

[Jane Manchun Wong @wongmjane]

DeepSeek V4 “improved” the code and said nothing happened in Tiananmen Square on June 4, 1989

[Evan Liu @evan123liu]

@grepmoney Sounds like Chinglish

[Evan Liu @evan123liu]

@lexfridman Are you gonna interview CC Wei or Morris Chang???

[Yun-Ta Tsai @yunta_tsai]

Some say that with AI, you don’t need to learn anything.

But in reality, there’s no better time to learn everything.

[Evan Liu @evan123liu]

New Claude Code hack:
Use /goal and then ask it to use as many parallel agents as possible all connected to its Chrome Extension MCP to test out every feature, click every button and all the frontend routes in all the different screen sizes. And then use voice to transcribe 30+ different things you want it to do. Ask it to run for hours and then do test and research extensively. Make sure it follows the design rulebook. After that have it auto commit and push to the remote preview branch so that Github Workflow auto deploys this preview app and you can test the app running from the cloud right after that. And then make sure the agent you're chatting with always delegate tasks to subagents so as to not waste its context window. It's like a coworker you're working with.

[Evan Liu @evan123liu]

It's so easy to say that this AI boom is a bubble and then everyone just agrees with you.

It just goes back to the saying that "pessimists get to sound smart and optimists get to make money."