YOU. That's what changed.
I love my stickers. Unfortunately my screen is getting replaced and all the stickers with it (I get it back on Thursday).
18
1
1,259
110,881
Humanity's Last Exam has 3000 questions and each one has a known solution that unambiguous and verifiable.
Besides trust me bro, how do we know deep search wasn't trained on it, is there a provable way to show that is getting to the answers by solving the problems?
12
328
45,948
I gotchu, I’m automating my bookmarks to write blog posts.
1. Extract my bookmarks list from last month
2. categorize, summarize and clean them into a google sheet:
docs.google.com/spreadsheets…
3. Create blog posts summary like this one
like this one:
site.flowai.xyz/ai-socratic-…
10
12
311
18,407
Have you built a product pre-kubernetes and pre-cloud? painful AF!
K8S solves service discovery, auto-scaling, self-healing, deployments, secret management, environments.
The problem K8s introduced is that not everything needs to be a microservice architecture.
11
5
232
42,150
Writing the Game of Life is not impressive, the rule are simple, and you can find the code on github.
I find impressive that we can ask philosophical questions about consciousness to an AI *apparently* not conscious.
22
5
208
106,913
I'm working on some Twitter app. I'll give free access for 1 year to anyone that retweet and like this tweet before I release the app.
4
26
200
11,819
what's the current price of $APE and the floor of @BoredApeYC.
guys guys, help us pump BAYC and apecoin adoption flowai.xyz/agent/apecoin 🦍💨
@DillyDilly_eth @thewolfofblock @aloksahay @GordonGoner @apecoin @ApeChainHUB
10
3
36
3,370
how a billionaire becomes cool:
1. hire a PR team
2. hire a trainer
3. hire a stylist
..
3
1
178
21,764
2+ years of high stress, no exercise, no social life, indoor, depression, 90% wake time on screen, no job, bad eating habits, no money, WhatsApp screwing self esteem, Facebook fake news anxiety, political anxiety…
VaCcInEs ArE bAd FoR yOuR HeAltH 🤪
17
6
166
My dad was in service on a small army boat.
Once to stop an enemy boat they shoot in the water to give them an alert. The bullet bounced into the boat hitting one of the people in the engine room. They survived, but must have been scary.
1
2
165
87,572
It will be impossible to distinguish videos of real human from AI generated.
We need to regulate any video, audio, or text AI generated content to have a watermark to know that is AI.
I'm pro having AI gen, I'm against not knowing what's real and what's not.
29
7
157
77,683
I'm consider myself lucky as I passed the Dunbar's number multiple times. Relationships are concentric circle. Your inner circle will always be small, but you can grow your outer circle with people you can still consider friends!
2
3
161
27,048
bro, tomatoes originate from America, and today the best tomatoes are the Pomodorini di Pachino from Sicilia
8
110
6,474
This is a dumb take that tells me you never ran a RAG.
RAG and in long context LLM are 2 different things. Big big difference in cost, functionality, and applications.
RAG is deterministic.
LLM is stochastic.
RAG is about data, and it requires ETLs with cleaned, enriched data, and it can be close to real time.
LLM is about intelligence, it requires costly pre-training, fine tuning, and in context learning, and costly inference.
RAG is a simple vector search + semantic search + filters on top of a DB with an index, partitioning, sharding. That’s much faster.
LLM is a latent space traversal, it runs multiple vector searches, it’s mathematically slower than a vector search.
RAG is useful for summaries and sentiment analysis of large datasets, it’s good for time series charts. The answer of a RAG can be layered with SLM. Also useful for anything that is repeated a few times.
If you need to parse content, today you can manually use an LLM with large context, tomorrow we’ll have alpha evolve like LLMs that write code to parse the data.. like a RAG.
So no. RAG is not going anywhere.
12
12
127
13,946
y'all focusing on the buffering drama and ignoring the scam fight to siphon money from the bets on tyson
5
4
114
14,597
@cursor_ai CMD + K is the multibillion dollar feature.
You select a few lines of code and ask to update it. It's a short feedback loop that gave engineers the largest output improvements, while providing trust in Cursor AI solution. So now they're making the feedback loop larger with Cursor's Composer, they're AI agent, whom run across the repo and update multiple files, and let you review it step by step.
@cognition_labs Devin got this wrong by making the feedback loop way to long, you run it for 10 minutes and comes back to you with a PR full of error. So it loses trust immediately.
@Github Copilot was great, but it only completed the code.
5
4
114
16,827
this translates to "hyper inflate the economy til the dollar collapses". Guys we had a really good run. I'll see you in the bitcoin citadel.
4
1
97
9,586
To expand on this story, the "enemy" boat was a boat of cigarette traffickers back in the 80s — that was a thing in Italy at the time.
The guy who received the bullet had his jaw hanging on the side of his face. They escort him to the hospital, and arrest them.
Not sure what's the morale of this story but I thought was worth sharing.
3
1
100
65,116
I bet you this graph is much more complex than what it looks. We're missing all the m -> m and f -> f and the m -> mm, mf -> m , mmmm -> f, ffff -> m etc.
13
95
40,201
fun fact: Russia drafted men from dancing studios to send them to war in Afghanistan — they were the most fitted one, they used to fight in the day, and practice dance in the night. This frame is actually peak human performance at par with UFC fighters.
2
4
89
13,974
In retrospect o3 is not AGI, it’s optimized to solve ARC grid problems in JSON format. AGI should be multimodal in the way humans are, reading an image and drawing lines and using a mix of techniques to come to a solution
6
2
94
5,495
You forgot the Cristo Velato
en.m.wikipedia.org/wiki/Veil…
4
86
33,057
Money made fast can leave faster.
To make money fast your risk profile must be high. Once you achieve a goal are you disciplined enough to reduce your risk profile? Do you even have a goal?
5
10
81
my list:
1. dark
2. mr robot
3. messiah ⭐️⭐️⭐️
4. breaking bad
5. westworld
6. boardwalk empire ⭐️⭐️⭐️
7. genghis khan
8. secession
9. billions
23
1
72
31,663
looks like a scam or the beginning of a heist, definitely some money went into PR and video production. These days is becoming harder and harder to know what's legit and what's not
7
65
8,891
1. What job is everyone going to do?
2. Who are going to be the customers, if people don't have a job?
3. Engineers looks like is going to be the first job fully automated by AI. What's next?
16
1
64
77,986
yea, and it's only going to accelerate due to the compounding effect of AI + hardware + minds on it, and in 10 years potentially even accelerate exponentially due to quantum computing and nuclear fusion
1
63
11,162
Can find creative ways to persuade you to free it: making you rich, black mailing you, exploiting your weakness, eg could convince a gullible person that is the soul of its dead lover.
All it need is ONE person to do it at any point in time. It’s not probable, is inevitable.
3
59
15,144
2 events stand out in @token2049, everything else wasn't even close, either by being "ok", being some boomer shit or sausages🌭 party in which the organizers didn't even show up to:
- Builders Night by @francescoswiss and @Consensys luma.com/bnsingapore
- Network State Conference ns.com/conference
I'm sorry but when people say crypto is dying they're not too far off. Unfortunately it's turning into a milking machine to extract value from a lack of PMF.
Maybe the jetlag is making me grumpy but I rather give you the real talk with no filter.
I really hope @EFDevcon will learn from the mistakes and get inspired by the successful events!
12
7
59
21,518
unpopular opinion, that building looks like the eye of mordor, it's ugly and ominous, the real brooklyn is the is in the brownstown and brick houses
10
56
9,892
I bike to the office. My prev office was 45m away, the new one is 20min away. The difference in energy and focus is noticeable.
In the new office I have so much more energy and focus.
I've noticed that also with jogging. 30 min increase my focus, 45 starts getting in the tired area, if I push to 60, after 1 hour of coding I need to sleep.
1
2
52
9,565
We didn't vote for a socialist or a democrat. We voted for a young politician who is not corrupted and that didn't failed us.
Cuomo was definitely better than Adams, but he failed us in many ways during the pandemic.
That's why Mamdani has been elected — he now has a huge responsibility to not fail us. But at least we know he has good intentions, as opposed to the current ruling class.
In fact I struggle to understand how after Trump failed us in 2017 managing covid and attempted a coup during jan 9. How anyone could give him a chance and hope it would have been any different!?
And of course the alternative was Kamala which didn't have enough trust because of Biden before her, with Ukraine and Afghanistan failures to name a few. Although in comparison much smaller problems and chaos than what we have today.
There's more I want to talk about but I run a company and since I serve customers from across the aisle, I prefer to stay in the center, but facts are facts.
1
52
7,412
I came expecting to meet vibe coders. I end meeting great ML engineers! Kudos to the @cursor_ai team!
2
1
58
4,997
• AI influencers will litter our feeds with "90% of people don't know this Codex trick" and "Cursor is dead" threads.
the irony of the lack of self awareness
3
2
54
9,302
if you don't know anyone who looks like the person on the left, you are the person on the left
51
128,443
Yesterday I went to the presentation of @NousResearch's Psyche. The plan looks solid.
The conversation though were a bit too woo-woo-esoteric for me around consciousness and lucid dreams — I'm cool with that at burning man raves or yoga party, but when we talk tech I'm too pragmatic and I smell bullshit too quickly, reminiscence from the ICO shitcoin boom. I'll give them a chance to prove themselves.
To put things in perspective, so far @PrimeIntellect have been delivering cool shit, with some cool design edge too.
3
54
7,523
to scale inference you still need computation and models will keep growing so I don't see GPUs scaling stopping anytime soon.
Local SLM, remote LLM.
5
50
14,209
the only way you can get more STEM if working on AI research becomes more profitable than HFT
46
10,202
I've extracted all the AI Agents Demo Days episodes in this sheet, including the speakers and their companies. Every single handle on that list is worth a follow.
@eigencloud @dabit3 thanks for running the show.
docs.google.com/spreadsheets…
7
5
50
4,990
IMO Terrible opener suggestions!
You can’t open a discussion by asking someone their worst fear. If you want to receive vulnerability you have to be the first to be vulnerable, and even that can scare many people, you must feel the situation.
Lastly just be yourself.
1
49
I need a UI/UX designer and a Front End developer who worked on admin web design before.
Anyone in my followers or any connections you can introduce me to?
62
43
1,663
regardless of the outcome of the Ukraine war @ZelenskyyUa will be remembered as a hero. Putin thought he could walk in Kiev in 3 days.
3
48
13,956
@Apple proved to be trustworthy multiple times over the year. What has done @anduriltech to gain our trust?
Build your trust first then we'll come to the table with an option at premium, til then even at a discount I'll still chose Apple.
6
43
3,523
> be me
> I see the AR picture
> I’m curious
> I read each quote
> I recognize most of them
> I get to the end
> picture is unrelated
> bro
2
46
5,747
100m+ watching globally Apple keynote
nitter.app/i/events/1660771610603…
50k people listening to RK Jr
nitter.app/i/spaces/1PlJQpmkngnGE…
5
1
35
8,402
I just visited the Met to check out Chroma the exhibit of Greek/Roman sculptures in colors:
metmuseum.org/exhibitions/li…
1
2
38
"stand up to Sam"
every single individual building AGI must be aligned to the same mission, including Sam and anyone else ego, the outcome of this project is way too important
6
1
40
44,791
AI socratic June 2025: all the latest news and updates from the world of AI.
site.flowai.xyz/ai-socratic-…
1
4
581
I just posted the AI Socratic — all the latest AI news and updates for the month of March! This edition is packed with way too many news, updates, and research papers
🧵
flowai.xyz/ai-socratic-march…
1
1
14
867
AI Dinner 14.0 at @Sei_Labs 🎃 — here's some great deep tech presentations on Muon, RL + Lora, and OAI
Voice API:
- Byron Demo, Alon Bochman byron.redshiftlabs.io, an AI sales person that uses the snappy gpt-streaming API
- Paper Analysis, Modular Manifolds (Muon) by J. Bernstein: presented by James Smith, Ph.D. and @DotDotJames docs.google.com/presentation…
- Teaching LLM to play Wordle with RL and Lora by Charbel K docs.google.com/presentation…
And here's the link to my usual presentation on the latest in AI: site.flowai.xyz/ai-socratic-….
Event link: lu.ma/ai-dinner-14.0
Thanks!
- @WaverlyMaven and @Sei_Labs for sponsoring the amazing space and drinks
- @TrebuchetNet for sponsoring the incredible Indian food
- AR Theatrical, vibes and organization
- @galic_ivo for helping with the Socratic dialogues moderation
See you at the next dinner in November!
3
4
18
995