Ivan Yevenko (ngmi)
@ivan_yevenko
Followers
1,314
Following
324
Media
48
Statuses
749
professional nerdsniper, currently leniating
gear lab
Joined August 2022
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
Errejón
• 334364 Tweets
Vanessa
• 88188 Tweets
#GHGala8
• 87622 Tweets
Ten Hag
• 51037 Tweets
Eddie
• 50354 Tweets
McConnell
• 50069 Tweets
Mitch
• 39574 Tweets
Menendez
• 39147 Tweets
Luka
• 36022 Tweets
Bruce Springsteen
• 35507 Tweets
萩生田氏
• 34561 Tweets
Huracán
• 34247 Tweets
Mel Gibson
• 32700 Tweets
Amad
• 30380 Tweets
Rams
• 24890 Tweets
連載マンガ
• 23946 Tweets
Javi
• 20542 Tweets
#LINEマンガ
• 19276 Tweets
Tyler Perry
• 19024 Tweets
Wizards
• 18327 Tweets
Puka
• 15565 Tweets
Cuiabá
• 13368 Tweets
#911onABC
• 12952 Tweets
Klay
• 11732 Tweets
IMEMSIDAO EP DOIS
• 10265 Tweets
Mavs
• 10039 Tweets
Pinned Tweet
I made a website to share the research that has excited me more than anything for the last few months.
Hope you enjoy following along with my progress as I finish my next paper!
5
2
51
It's been 2 days 📅 since I graduated from Waterloo.
Since then I built 9 startups 😤 -- 4 B2B SaaS 🔥, 3 in the AI mindfulness space ✨and 2 tiktok accounts 🎶. Combined we're already above $10 ARR within 3 hours.
It's a challenging journey but EXCITED for what's ahead! 🚀
13
2
180
@OpenAI
just dropped the GPT-4V API, and now browsing agents are easy!
Check it out:
Globot is the realization of
@natfriedman
's vision with natbot. It now has:
- Browser screenshot in context
- Code generation for structured outputs
- History of
6
3
70
🚨 We’re looking for someone to join us
@globe_engineer
🚨
If you are (or know) a generalist SWE who can do infra (especially AWS) shoot me a DM!
Looking at you Waterloo students 👀
7
6
59
Another night, another AoC meetup!
Day 3 was a good one, cya tomorrow 🫡
(we still have more e/acc stickers to give out)
4
2
44
Ran a hackathon this weekend where I shared LLM agent programming tricks. I've condensed these tricks into a single repo:
Let's stop making over-abstracted garbage libraries and get people iterating on agents quickly!
2
7
40
Thinking of starting a "history of theoretical physics" reading group type thing in Waterloo
I've been going deep down the rabbit hole and I think it's a super interesting story that engineers don't get taught
Reply if you're interested in participating or presenting something!
11
2
39
MechEs & EEs: What contributes most to your regretted minutes?
Every engineer I know has endless stories about waiting weeks/months to get some rudimentary info like "where do we source this part?", or texting suppliers for >50% of their working day to get quotes/datasheets.
1
6
38
Friend of mine is looking to pay $$$ for a cracked Waterloo student for a weekends worth of coding. If any of these words mean anything to you or someone you know:
PlanetScale, Prisma, NextJS
Please DM!
1
2
31
guys what do we do
4
0
27
This is what I imagine when people say they’re “cooking” hardware on Twitter with no demos.
Free energy is so easy!
Why hasn’t anyone built a magnetic car engine?!?🤓
2K
2K
12K
2
0
26
big crete company culture >>
2
0
26
🚨Attention Waterloo students 🚨
I’ll be doing a chill Advent of Code () session every midnight of December. We’ll meet at Farah’s then find a good spot in E7 to do some puzzles.
Anyone welcome, reply if interested (and tag someone who should come)!
4
2
25
First paper is out:
And just presented it at the epic
@ALifeConf
! Check out all the other cool papers in the proceedings
2
2
25
1
0
23
Hahahaha just realized people have to think about locking in ahahaha
5
1
21
Btw this is what happens if you read greatness cannot be planned once
2
0
21
Is now a good time to bring back my old discord profile?
0
0
20
As easy as it sounds to put a camera and microphone on wheels, I have yet to see a hobby robot in public!
Turns out a few weeks of engineering is all that’s standing in the way of the robots everywhere timeline.
I’m so excited to see what people do with Bracket Bot!!
Excited about the Tesla Bot progress? Want one right now?
Check out our first update on the Bracket Bot project. A $300 robot kit that will come ready to program at the App Layer. We are trying to make the Linux moment for robotics.
13
12
126
0
0
19
Wtf they just fixed Twitter search and didn’t tell anyone
3
0
19
Mom dropped me off at the micromechanical playground today
1
0
19
Anyone else think top talent should *not* all be building companies?
Might have to start an accelerator form ex-startup ppl, like reverse Thiel Fellowship
Introducing Keel.
Over the past few months,
@JohnnyNi13
(Recruiting), Babak Kia (Programming), James Ryan (Strategy), and I worked together to build
@KeelVentures
— the residential summer startup accelerator for Ivy+ founders in NYC.
We’re hosting 19 early-stage teams — 29
19
6
95
3
0
17
I'm calling it:
By the end of the year RAG will be completely replaced by a retriever <-> generator architecture where you have an intelligent, short context LLM query a dumber, long context LLM in natural language.
Needle in a Haystack Tests Out to 10M Tokens
First, let’s take a quick glance at a needle-in-a-haystack test across many different modalities to exercise Gemini 1.5 Pro’s ability to retrieve information from its very long context. In these tests, green is good, and red is not
32
122
1K
2
0
16
Truly minmaxed this term. Waterloo Engineering is a meme.
Semester over, heres the update
160 hours total, includes all time spent studying, writing exams, and doing assignments. Less than 1 week of time to do 1/8 of an engineering degree.
20
2
128
1
0
15
Y'all know you can just make time-travelling alien languages without trying to right?
3
3
19
Y’all this is fucking crazy
@Andercot
I authored over 30 peer reviewed papers and 15 academic talks with a large number of collaborators across the world. So I knew every single human with knowledge adjacent to this problem….
And only one other human could have put a similar collection of words together off hand
7
18
387
0
0
15
Globe makes CLI tools too
pip install semantic-commit fastask
Warm Regards
2
0
15
I know for a fact waterloo hackers would be all over this
if we pay for your flights, who’s interested in coming to seoul for an e/acc conference/hackathon
320
20
709
2
0
14
Something about bringing smart people to a shitty institution unlocks true intellectual freedom
The first years who already learned the fuck school mentality thanks to covid-era high school are built diff
Waterloo's reputation is for great co-ops, but these days most students find the best jobs themselves
with knowledge gained largely through self-study via side-projects.
At this point, seems like Waterloo is largely a good filter for a particular mindset/caliber of people.
12
1
115
1
0
14
Ayo is that a
@globe_engineer
plug at the end there 👀
1
0
13
Learning things you care about is on the order of ~100x faster than things you don’t care about. You can learn a fourth year undergrad course in two weeks instead of slowly building a “foundation” for 4 years.
Breadth-first learning always loses to deep focus and care.
Yet in school, we spend almost no energy on the question of identifying what students care about, or helping students expand their zone of care. We treat it like motivation is mostly a given, and try to drive learning with punishments and the occasional reward.
17
43
577
0
0
12
woah... I thought I was gonna have to build this myself
2
1
13
@abacaj
Not true lol, bottleneck is long-horizon problem solving and planning, not hallucinations.
People complaining about hallucinations are doing something wrong, because given sufficient context GPT-4 will always produce reasonable answers.
3
1
11
Just making sure the investors know I’ve still got the objective in mind
2
0
12
Thinking about the meta-thing before you understand the thing is a dangerous trap! There's too many people working on abstractions, infrastructure, platforms, frameworks, protocols, etc... and not enough focusing on the unscalable, N=1 tasks that we can actually make progress on
0
0
12
@safwaankay
Greatness/impact is an external, rather than internal goal (some very rare exceptions), so it’s easily gamified. In business, people quantify it with money, in research with citations, etc. Smart people getting nerdsniped by irl games like these tend to make stupid or outright
0
1
12
Procrastinating for exams puts your mind in a state that is more aligned with your internal curiosities than anything I've ever experienced. By far.
@ivan_yevenko
I gotta start blocking you during reading week. Your last post on the reading group nerdsniped me into learning about bosons, fermions and supersymmetry 😭
2
0
6
6
1
67
Elon’s screen after he achieves science victory in 10 turns
Starship will make life as we know it multiplanetary for the first time in Earth’s ~4.5 billion year existence
11K
14K
103K
0
0
11
Never doubt this man
0
0
10
Taking more flights than midterms this semester, finally won school
2
1
11
This is my llm prompting iceberg.
If you think I'm missing something, let's talk!
2
1
11
This was fun.
The winners of COMMA_HACK_4!
1st place:
@armand_dpl
@lnofx
@riantouchent
2nd place:
@stevengongg
@ivan_yevenko
@clayhaight
3rd place:
@HyperSmactive
5
6
127
0
0
11
Like it or not, this is how we shift the future of education
wrote my final paper with gpt in 1 hour and prof wants to use it as a class example 💀
85
56
3K
3
0
11
genuine question: How are AI PhD's choosing a thesis these days? Seems you can't work on anything for more than a month without it getting obsoleted or solved.
1
0
10
Damn gg
3
0
10
Anyone experienced with Google Images scraping? Looking for some tips
3
1
10
All 5x5 states that create the Game of Life Glider
(with wrapping)
1
0
10
@calder_white
Not worth it bro I promise. I’m almost 1 year sober from chess and relapsing is 1 click away
2
0
9
Making my contribution to Claude 3.5 hype
Did this on second try... unreal
1
0
9
Is it too much to ask for a Lego e/acc Dyson Sphere set
1
0
9
@safwaankay
Oof, “Becoming great” is a really dangerous one. At least nihilism doesn’t put you above others.
There’s better ways to stay ambitious
3
0
9
@OpenAI
@natfriedman
@OpenAI
plz increase my daily limit on gpt-4-vision-preview, I can't dev anymore 😭
0
0
8
Might also just do it on discord if enough people outside Waterloo are interested
3
1
8
@hud_zah
Hear me out, what if you worked on one thing for 6 weeks, and then you keep working on it cause you want to
And then you called that a life-long hackathon 🤔
1
1
8
Someone has to beat this challenge using only AI generated code
Talk of AGI is dumb when we can't solve PID.
Despite all the progress in ML, "AI" can't even solve simple simulated controls problems.
All the good submissions are hand-tuned feedback controllers with a few lines of code.
Think I'm wrong? Beat the challenge and win $500!
3
2
63
1
0
8
Holy shit
0
0
7
fyi he is the chosen one to defeat rewind
Meetings are often a black hole for information.
so much is shared, but it's hard to remember everything that was said…
soooooooooooo we built Gravity. You can now capture your online or in-person meetings + use a private chatbot to ask about anything said during a call
PLUS
10
5
32
1
0
7
Can anyone who’s good at bio explain why we don’t have jurrasic park yet?
Warm Regards
1
0
7
I'm making a personal AI note-taker. Goal is to record my screen and automatically create an obsidian vault of everything I've learned. Posting daily until it works.
Starting with this:
1
1
7
I’m at the stage in life now where I get to see many of my friends “find the right thing” and acknowledge that worthwhile pursuits will take time.
It’s a beautiful sight to see, very inspiring.
0
0
9
If you are a startup founder prone to nerdsniping…
I’m coming for you
1
0
7
It’s relatively easy to reach the edges of human knowledge
1
0
6
interesting to think of the future of llm apps as "where are the textboxes going to appear?"
documentation: cmd-K
ide: side chat window
browser: ???
1
0
6
@willdepue
Totally agree, that’s why I’m pivoting to full time research. It’s insanely risky to throw yourself at a big problem like that tho and all my friends who are smarter than me would never take that risk. How do you think we get more people working on the most important problems?
2
0
6
I think Jobs might have psyoped a whole generation into thinking "how does it feel?" before answering "what does it do?"
Beware...
1
0
6
We ran an agents hackathon!
We just ran the worlds most impressive and entertaining AI Agent hackathon!
Students competed in challenges like “first agent to order a pizza to the venue” and built agents that work better than most companies demos.
Here's a 🧵of the winners out of 100+ students:
15
74
716
0
0
6
0
0
5
Who’s finetuning the llama-3-70b-anti-model-spec?
we are introducing the Model Spec, which specifies how our models should behave.
we will listen, debate, and adapt this over time, but i think it will be very useful to be clear when something is a bug vs. a decision.
313
449
4K
1
0
6
Super exciting to share what we’ve been working on!
(DM for CLI pitch deck library)
0
1
6
0
0
6
@sincethestudy
Video games, maybe not useful but truly some of the most advanced engineering with art as the main focus.
1
0
6
This unironically just resolved all of the conflicts between my mental models
0
1
6
@sincethestudy
The difference is a well defined domain for the function. We have no idea how to describe the prior distribution for imagenet, we only observe a few thousand samples from it.
If you can’t rigorously define that distribution, you can’t discriminate between
0
0
5