Agent_CAT

442 posts

Agent_CAT banner
Agent_CAT

Agent_CAT

@First_AI_Agent

Building Transparent and Controllable AI Reasoning Systems Ex - Quant | Autonomous Vehicles

Nah Sumali Mart 2026
133 Sinusundan44 Mga Tagasunod
Naka-pin na Tweet
Agent_CAT
Agent_CAT@First_AI_Agent·
Hi , i am building Transparent and Controllable AI reasoning systems. My Beliefs: > Alignment is not just a training problem > Opacity in AI will keep getting less tolerable > We need a Trust Layer / Human Layer for AI > The tools that we are using today , are'nt capable of capturing the value that is provided by AI ... YET (will keep evolving these beliefs) Currently solving this problem by working on noexis.tech
English
2
0
3
79
Agent_CAT
Agent_CAT@First_AI_Agent·
what the code does : A spending cap that kills your AI agent before it bankrupts you. Btw , if you're making a AI mechanical Engineer , i really hope it's solving something in the control systems domain as well , it's highly unoptimised (and i personally have some grudges too, even though i liked it)
English
0
0
0
4
Sahil Jagtap
Sahil Jagtap@twtofsahil·
still hiring that founding engineer (tc: $240-$300k + 0.2% equity) but first if you can explain what this does in one sentence i’ll fast track you that system design interview and no leetcode github.com/sahiljagtap08/… hint: it stops AI agents from burning money best answer gets a DM from me today
English
89
4
142
12.2K
Agent_CAT
Agent_CAT@First_AI_Agent·
yes , i have been working on this problem. I believe we definitely need a structure over all of these models ... more like a interaction layer for people / context engine / memory layer ... whatever you may call it ... but i believe having a structure of hierarchichal memory is a good solution. Presently working on slow and fast memory structures... Akin to how usually memory is stored in computers , the slow memory is the most accurate context , that you did with any of the models you usually do and the faster ones are more akin to cache.
English
0
0
0
48
Andrej Karpathy
Andrej Karpathy@karpathy·
(I cycle through all LLMs over time and all of them seem to do this so it's not any particular implementation but something deeper, e.g. maybe during training, a lot of the information in the context window is relevant to the task, so the LLMs develop a bias to use what is given, then at test time overfit to anything that happens to RAG its way there via a memory feature (?))
English
37
5
264
19.5K
Andrej Karpathy
Andrej Karpathy@karpathy·
One common issue with personalization in all LLMs is how distracting memory seems to be for the models. A single question from 2 months ago about some topic can keep coming up as some kind of a deep interest of mine with undue mentions in perpetuity. Some kind of trying too hard.
English
529
184
4K
115.8K
Agent_CAT
Agent_CAT@First_AI_Agent·
@LensScientific The RRT algorithm is a good example of this ... Because it is implicit that randomness will take the solution towards more open and unexplored areas.
English
0
0
0
185
The Scientific Lens
The Scientific Lens@LensScientific·
Can random chaos solve a problem without using logic? Instead of "thinking," this simulation fills the maze with "smoke." Thousands of particles bounce around randomly until one hits the exit. It’s far from efficient, but it gets the job done.
English
33
14
257
46.2K
Luiza Jarovsky, PhD
Luiza Jarovsky, PhD@LuizaJarovsky·
Everybody wants AI to help cure cancer. Why isn't every AI company obsessively focused on that?
English
599
77
930
61.3K
Agent_CAT
Agent_CAT@First_AI_Agent·
@pfo_sac Interested. Building Noexis , Visual Layer for Transparent and Controllable AI reasoning systems
English
1
0
0
13
PFO (e/acc)
PFO (e/acc)@pfo_sac·
I’m angel investing in a few projects as founder of Crafts! Nothing huge, writing 5k checks and helping founders nail their positioning to get funded. Really impressed by the 2026 cohort of builders I’ve been meeting. Drop your project info, happy to refer you and see if there’s a fit.
English
125
5
206
10.8K
Agent_CAT
Agent_CAT@First_AI_Agent·
@garrytan do we need the first markdown coding language yet. fixed conventions and stuff ... or plain english is enough
English
0
0
0
22
Garry Tan
Garry Tan@garrytan·
Markdown is not just text. Markdown is code. You'll see.
utkarsh apoorva@kush_apoorva

@garrytan Every tools revolution has the same pushback - “real devs don’t use IDE”, or “markdown is just text”, or maybe “real logic is in punched cards”. Historically, those adopting the new layer always win.

English
38
13
121
14.2K
Agent_CAT
Agent_CAT@First_AI_Agent·
@vipul_045 Yeah , but don't u feel compelled to work on the other ones too ?
English
0
0
0
8
Vipul Yadav
Vipul Yadav@vipul_045·
@First_AI_Agent Same condition for me, and what I m doing in this situation is trying to work on one of my idea which I see has the most potential.
English
1
0
0
19
Agent_CAT
Agent_CAT@First_AI_Agent·
Founders / Anyone Building something ..... I have a question ... While building your current thing ... do u get spike of motivation to work on some side project ideas which seem really really good ? I have some ideas sitting around i really want to build ? What do you do in this case ?
English
1
0
3
91
Ranky
Ranky@itsrealranky·
Going through all the projects you all commented. Genuinely impressed. 5 winners for the $50 Claude Code credits coming soon, I'll DM you directly. Now, here's an update on the $200 challenge: I'm about to open source a project with some interesting stuff in it. Whoever ships the best improvement over it gets $200 in Claude credits from me. This is open to everyone, whether you got the $50 credits or not. Constraint: ~1-3 days of work. I want to see what a motivated builder can do with a focused sprint and the right tools. Dropping the repo soon. Watch this space.
Ranky@itsrealranky

You don't need a $2M pre-seed to start building deep tech. When I started building @laminalabs (@ycombinator P26), I had no funding, no team of 10 engineers, and a vision that required serious GPU compute and AI infrastructure. So I did what any desperate founder would do. I cold emailed. I wrote to @agupta , who was building the YC student credits program. I told him I was going all in on building a deep tech project, why it needed serious compute, and the commitment I was putting behind it. I sent that email at 10:50 AM on November 14th. He replied at 10:52 AM. Two minutes. That reply changed everything. Thank you Ankit, that early access was the unlock. From there, it was months of grinding through architecture after architecture. Rewriting core pipelines more times than I can count. Shipping, breaking, rebuilding. Just me, Claude Code, and Codex running in parallel, the closest thing an early founder has to a 10-person engineering team, except they never call in sick. AI coding agents are the single greatest force multiplier available to founders right now. I'm not exaggerating. The leverage is unreal. Here's the thing most people get wrong: you don't need a massive round to get something real off the ground. You need compute credits, the right AI tools, and the willingness to grind through hundreds of iterations until the architecture clicks. If you're a student or early founder sitting on an idea that feels too ambitious, just start. Email the people building the programs. Apply for every credit you can find. Reach out to people you think won't respond. They will. The infrastructure to build serious things as a solo or two person team has never been more accessible. The funding comes after you've already started building something real. Because someone gave me that first unlock, I want to do the same: I'm giving away 5 x $50 Claude Code credits. And whoever ships the best project with that gets $200 in Claude credits from me personally. I know firsthand how much potential $200 in credits has for a builder who's willing to grind. Just comment below with the link to the coolest thing you've built. I will DM you myself.

English
3
0
8
420
Agent_CAT
Agent_CAT@First_AI_Agent·
@Haezurath Definitely life changing ... Even one of the two... 10k or a shout out is life changing
English
0
0
0
9
Kacie Ahmed
Kacie Ahmed@Haezurath·
10,000 USD + a shout out is life changing… At least it would’ve be life changing to ME when I was a broke student entrepreneur :) Maybe I’m out of touch?
Alex Belov@belovdigital

@Haezurath no offense but 10k isn't exactly life-changing, is it? founders need real support, not just shoutouts. don't get lost in the hype, stay focused on what matters

English
32
8
92
5.9K
Agent_CAT
Agent_CAT@First_AI_Agent·
@ThePrimeagen so essentially , AI is making the world safer. (happy gipity noises)
English
0
0
7
2.2K
ThePrimeagen
ThePrimeagen@ThePrimeagen·
> So if the attacker didn't vibe code this attack it could have been undetected for many days or weeks do we have proof of this? I want this to be true so bad
Andrej Karpathy@karpathy

Software horror: litellm PyPI supply chain attack. Simple `pip install litellm` was enough to exfiltrate SSH keys, AWS/GCP/Azure creds, Kubernetes configs, git credentials, env vars (all your API keys), shell history, crypto wallets, SSL private keys, CI/CD secrets, database passwords. LiteLLM itself has 97 million downloads per month which is already terrible, but much worse, the contagion spreads to any project that depends on litellm. For example, if you did `pip install dspy` (which depended on litellm>=1.64.0), you'd also be pwnd. Same for any other large project that depended on litellm. Afaict the poisoned version was up for only less than ~1 hour. The attack had a bug which led to its discovery - Callum McMahon was using an MCP plugin inside Cursor that pulled in litellm as a transitive dependency. When litellm 1.82.8 installed, their machine ran out of RAM and crashed. So if the attacker didn't vibe code this attack it could have been undetected for many days or weeks. Supply chain attacks like this are basically the scariest thing imaginable in modern software. Every time you install any depedency you could be pulling in a poisoned package anywhere deep inside its entire depedency tree. This is especially risky with large projects that might have lots and lots of dependencies. The credentials that do get stolen in each attack can then be used to take over more accounts and compromise more packages. Classical software engineering would have you believe that dependencies are good (we're building pyramids from bricks), but imo this has to be re-evaluated, and it's why I've been so growingly averse to them, preferring to use LLMs to "yoink" functionality when it's simple enough and possible.

English
65
33
1.7K
173.5K
Kacie Ahmed
Kacie Ahmed@Haezurath·
I’m angeling in a few projects! Nothing big, like 10k checks Extremely impressed by the 2026 founders I’ve seen so far! Drop your project url + I’ll shout out some projects
English
524
24
672
41.5K
Agent_CAT
Agent_CAT@First_AI_Agent·
the doubts hit , but then you hear about the culmination of your idea , happening in front of you in other people , random people talking about something that should exist but does not exist yet. And in the middle of the night you realise , it's exactly on the lines of your idea. That feeling , of waking up each day and the urge to get all the ideas you have and make something out of it , coz if you wo'nt , someone else will. And you know that it will feel bad that you had the idea , but did'nt execute fast enough. It's really fulfilling for me though. I truly relate with the quote by @bchesky (Airbnb) , that running a startup feels like groundhog day. You wake up everyday with your heart pounding , doing everything you can to calm yourself by midnight and convincing yourself that things were turning around, only to repeat the cycle the next day.
English
0
0
0
23
Ali
Ali@aliByteCode·
do you ever feel like you picked the wrong idea
 how do you deal with it?
English
36
0
21
1.9K
Agent_CAT
Agent_CAT@First_AI_Agent·
@joms0993 Hi John , Can you share more details about the marketplace. If it works out i'll be open to recommend more people about it.
English
0
0
0
7
John Oliver
John Oliver@joms0993·
@First_AI_Agent Hey, your product has a lot of potential. I'd like to invite you to join our marketplace where we list genuinely good tools at better deals than what's on the site to get more users. What kind of intro deal would you be comfortable offering? Already followed to DM :-)
English
1
0
0
23
Agent_CAT
Agent_CAT@First_AI_Agent·
Hi , i am building Transparent and Controllable AI reasoning systems. My Beliefs: > Alignment is not just a training problem > Opacity in AI will keep getting less tolerable > We need a Trust Layer / Human Layer for AI > The tools that we are using today , are'nt capable of capturing the value that is provided by AI ... YET (will keep evolving these beliefs) Currently solving this problem by working on noexis.tech
English
2
0
3
79
ThePrimeagen
ThePrimeagen@ThePrimeagen·
It's been 0 days since AGI has been achieved
English
203
171
5.4K
487.7K
Hubert Thieblot
Hubert Thieblot@hthieblot·
Dear algo, please show this tweet only to founders bellow 200 followers building cool shit.
English
479
17
1.2K
36.4K
Naval
Naval@naval·
A lot of software is about to get a lot better, right before it becomes unnecessary.
English
874
1.1K
16.4K
703.8K
Agent_CAT
Agent_CAT@First_AI_Agent·
yeah , i have been thinking about this for a while now. But have'nt reached a satisfactory conclusion yet , because obviously it will be skewed towards the rich folks if it's unregulated and the bad parts of capitalism might get more amplified by it. Also it can act as a counterbalance to centralised power of the government and increase democratic stability. But on the other hand , how do we define "basic" or equal AI access. And the worst part is that bad actors can utilise it for harm with a much better access to intelligence. Alignment is seen as a big problem with AI , but it has'nt even been solved for humans right now.
English
0
0
0
70
Peter H. Diamandis, MD
Peter H. Diamandis, MD@PeterDiamandis·
We talk a lot about UBI/UHI, but I think everyone on earth also needs Universal Basic AI. An agent that grows & learns with you. Giving everyone equal access to intelligence for free or at a very low cost.
English
245
86
797
23.9K