pratyush

14K posts

pratyush banner
pratyush

pratyush

@pratty_agi

building AGI one AI at a time @abundantlabs prev: @orangewoodlabs, @agentopsai host @agihouseindia

core เข้าร่วม Temmuz 2019
1.6K กำลังติดตาม2K ผู้ติดตาม
ทวีตที่ปักหมุด
pratyush
pratyush@pratty_agi·
Here’s are some of the experiments and observations I did as part of the initial testers on the locksmith game using within ARC-AGI-3 (my template is available in the repository) 🧵
ARC Prize@arcprize

Today, we're announcing a preview of ARC-AGI-3, the Interactive Reasoning Benchmark with the widest gap between easy for humans and hard for AI We’re releasing: * 3 games (environments) * $10K agent contest * AI agents API Starting scores - Frontier AI: 0%, Humans: 100%

English
1
4
52
29.6K
pratyush รีทวีตแล้ว
hari_haran
hari_haran@HariAyapps·
If you guys are working on anything interesting or cool in the intersection of self-learning agents, loop engineering, evals for long-running tasks, chip design, VLSI or semiconductor engineering, feel free to DM me or @naveen_venk A lot of cool people have been reaching out. Make sure you are one of them too xD
hari_haran@HariAyapps

We built the best chip design engine in the world!! 🥳🥳 @archgen_ is currently ranked #1 on the @WeAreHRT / Partcl Macro Placement Challenge 2026 leaderboard. Macro placement is one of the hardest problems in physical design. It involves placing large fixed-size blocks such as SRAMs, IPs, and analog macros on a chip floorplan while balancing, wirelength, density, congestion, routability, timing and constraints. After months of research and iteration our submission reached a verified rank-1 with an average proxy cost of 0.9507 across the IBM benchmark suite. @naveen_venk and @JishnuMada86596 burned the midnight oil to build an optimization flow that combined fast local repair, multi-start search, congestion-aware ranking, GPU-accelerated candidate generation and strict legality checks to reach the top spot. (detailed blog in the comments) Grateful to Madhusudan S, Abhishek Lal, and Anant Gulati for their valuable suggestions and inputs to help us overcome issues in EDA algorithms, traditional macro placement algorithms and GPU optimisation. If you are working on physical design and want to understand how AI, self learning agents, loops, and GPU-accelerated optimisation can improve your flows please feel to reach out to us. Thank you @Willschips, Vamshi Balanaga and the Partcl team for organising this competition. #PhysicalDesign #EDA #ChipDesign #VLSI #AIforEDA #Semiconductors #Placement #ArchGen #HardwareDesign

English
3
3
23
1.1K
Vinod Sharma
Vinod Sharma@VinodSharma10x·
I wake up at 4 am to work on Sucana before my day job starts. Weekdays and weekends. Consistently. For 1 year. That's luck.
Vinod Sharma tweet media
English
17
3
70
4.1K
pratyush
pratyush@pratty_agi·
@ariG23498 I hope this is not another event where only the known and famous are allowed. Someone here does this a lot to farm engagement, score high registrations and approve a few influencers only.
English
0
0
0
39
Aritra
Aritra@ariG23498·
We had to close registrations after a day. Thanks for the overwhelming amount of engagement.
Aritra tweet media
Aritra@ariG23498

[HF ML Club India] We are pleased to announce the IRL event in Bengaluru. We are partnering with @GoogleDeepMind and this takes place on the 15th of July 2026 (more details below). We will have someone from the GDM team talk about Gemma 4, and an hour long build session with Gemma and @huggingface. Great opportunity for folks to network and build. PS: We will not entertain any "where is the registration link" query. Seek and ye shall find.

English
6
0
43
2.9K
pratyush
pratyush@pratty_agi·
@vikhyatk > outlimit codex > go to events in the meantime > aura so high the event becomes a 3rd wheel > win everywhere
English
0
0
1
58
vik
vik@vikhyatk·
codex rate limit ran out. so i'm going to this openxla event in sunnyvale tomorrow
vik tweet mediavik tweet media
English
5
1
108
9.1K
i2cjak
i2cjak@i2cjak·
i just had an idea for possible defense applications, you could use drones instead of sending people to certain locations. perhaps you put large rocks on the drones and drop them on people. wait what about instead of rocks
Sacramento Sheriff@sacsheriff

In a nationwide first, a drone was successfully used to disarm an armed suspect, helping bring the incident to a safe resolution. Sacramento Sheriff’s Special Enforcement Detail (SWAT) had the residence surrounded. The known felon and Parolee-At-Large was seen earlier with a firearm. After not responding to negotiators, a drone was deployed inside the residence. Drone pilots located the suspect hiding in a corner of a garage. One of his hands was concealed under his body, and the other was holding a knife. He was obviously alive, but not responding. Was he pretending to be dead or simply lying in wait for deputies to approach? Rather than rush into a potentially deadly encounter, one drone pilot came up with an innovative solution. By attaching a powerful magnet to the drone, the pilot was able to safely remove the knife from the suspect’s hand before deputies moved in. It was an incredible display of creativity, skill, and precision by the drone pilot. It’s also worth noting that our drones are labeled as “military equipment” (even though anyone can purchase them at their local Walmart), but are really just another piece of technology helping deputies resolve dangerous situations safely. Their use protects both law enforcement personnel and suspects.

English
11
2
286
14.8K
pratyush
pratyush@pratty_agi·
@benln want to host one in New Delhi
English
0
0
0
40
Ben Lang
Ben Lang@benln·
Cursor community events in 200+ cities Reach out if your city's missing
English
125
40
675
55.5K
pratyush
pratyush@pratty_agi·
@thsottiaux Will have to test the limit of the guardrails since codex with gpt-5.5 on CVP still does not perform well on discovering vulnerabilities compared to Claude-code with opus on CVP but it’s good at generating exploits!
English
0
0
0
88
Tibo
Tibo@thsottiaux·
Let's Patch The Planet. Updates to codex security and a new GPT-5.5-Cyber. A day of celebration for cyber defense acceleration. openai.com/index/daybreak…
English
118
75
1.6K
112.3K
pratyush รีทวีตแล้ว
hari_haran
hari_haran@HariAyapps·
We built the best chip design engine in the world!! 🥳🥳 @archgen_ is currently ranked #1 on the @WeAreHRT / Partcl Macro Placement Challenge 2026 leaderboard. Macro placement is one of the hardest problems in physical design. It involves placing large fixed-size blocks such as SRAMs, IPs, and analog macros on a chip floorplan while balancing, wirelength, density, congestion, routability, timing and constraints. After months of research and iteration our submission reached a verified rank-1 with an average proxy cost of 0.9507 across the IBM benchmark suite. @naveen_venk and @JishnuMada86596 burned the midnight oil to build an optimization flow that combined fast local repair, multi-start search, congestion-aware ranking, GPU-accelerated candidate generation and strict legality checks to reach the top spot. (detailed blog in the comments) Grateful to Madhusudan S, Abhishek Lal, and Anant Gulati for their valuable suggestions and inputs to help us overcome issues in EDA algorithms, traditional macro placement algorithms and GPU optimisation. If you are working on physical design and want to understand how AI, self learning agents, loops, and GPU-accelerated optimisation can improve your flows please feel to reach out to us. Thank you @Willschips, Vamshi Balanaga and the Partcl team for organising this competition. #PhysicalDesign #EDA #ChipDesign #VLSI #AIforEDA #Semiconductors #Placement #ArchGen #HardwareDesign
hari_haran tweet mediahari_haran tweet media
GIF
hari_haran tweet media
English
53
39
267
20.6K
pratyush รีทวีตแล้ว
Greg Kamradt
Greg Kamradt@GregKamradt·
Today we announced a $1M donation from General Intuition @PimDeWitte, @EloiAlonso1, @AdamJelley2, and @micheli_vincent are directly supporting the open source ecosystem Funds will be used to drive ARC-AGI-4 and interactive benchmarks as a whole Onward and upward!
ARC Prize@arcprize

ARC Prize Foundation + General Intuition We’re grateful to announce that @gen_intuition has donated $1,001,337 to support the development of ARC-AGI-4 Thank you @PimDeWitte, @EloiAlonso1, @AdamJelley2, and @micheli_vincent for helping push open AGI benchmarking forward

English
6
7
42
10.7K
pratyush
pratyush@pratty_agi·
@rohanpaul_ai Easily hackable. Cybersecurity benchmarks lack proper verifiers especially those that use LLM-J.
English
0
0
0
98
Rohan Paul
Rohan Paul@rohanpaul_ai·
OpenAI’s new GPT-5.5-Cyber just beat Mythos 5 on CyberGym. CyberGym measures whether an agent can reproduce known software vulnerabilities, so this is quite a strong signal for defensive vulnerability analysis of models. OpenAI also launched a major push to use GPT-5.5-Cyber and human security teams to fix open source bugs before AI bug-hunting tools flood maintainers with low-quality reports. Vulnerability discovery is becoming much easier, so the scarce part is now remediation, which means confirming the bug, proving reachability, writing a fix, testing it, and giving humans enough evidence to merge safely. OpenAI’s initiative is to use GPT-5.5-Cyber as a defensive security worker inside Codex. It scans code, checks whether a vulnerability is real and reachable, writes a patch, tests the patch, and gives humans evidence to approve it. Daybreak is OpenAI’s new cybersecurity initiative to help trusted defenders find, verify, and patch vulnerable software much faster using AI. The new checkpoint of GPT-5.5-Cyber, are all part of the company's limited “Trusted Access for Cyber” program and do not involve a public release.
Rohan Paul tweet media
OpenAI@OpenAI

We’re expanding OpenAI Daybreak to help democratize patching vulnerable software at machine speed: - Codex Security plugin: find, validate, and fix vulnerabilities right inside Codex - The full version of GPT-5.5-Cyber model: a great model for trusted defenders - Cyber Partner Program: powering products built on top of our best cyber capabilities for leading security companies to secure the world's software - Patch the Planet: working with maintainers to secure critical open source projects openai.com/index/daybreak…

English
10
12
59
8.8K
Anirudh Amudan
Anirudh Amudan@anirudhamudan·
Absolutely love hanging out with tech bros outside of tech events, and got to love companies hosting such well organised, curated and tasteful events. There are cool companies in HSR, but no such cool events, all of them are in Indiranagar only 😭
Anirudh Amudan tweet mediaAnirudh Amudan tweet media
English
6
0
54
3.3K
pratyush รีทวีตแล้ว
Midnight Capital
Midnight Capital@Midnight_Captl·
On @chrisbarber’s show, I speak specifically about this concept, and why Cloud customers are going to increasingly care about what accelerator program they’re building on The Hyper’s complain about being locked into $NVDA, but their customers who use NVIDIA get to avoid being locked into them! (Which is also why they don’t love Nvidia) This is going to be an increasingly important factor as Cloud spend as a % of OpEx grows across most organizations
Jerry Capital@JerryCap

Why Apple is using $NVDA instead of Google TPUs. h/t @benthompson

English
3
6
33
9.2K
Aritra
Aritra@ariG23498·
@pratty_agi Let me tell you a secret, we have planned HF credits for the event. 😎🤫
English
1
0
1
21
pratyush
pratyush@pratty_agi·
The long awaited meet is here!
Aritra@ariG23498

[HF ML Club India] We are pleased to announce the IRL event in Bengaluru. We are partnering with @GoogleDeepMind and this takes place on the 15th of July 2026 (more details below). We will have someone from the GDM team talk about Gemma 4, and an hour long build session with Gemma and @huggingface. Great opportunity for folks to network and build. PS: We will not entertain any "where is the registration link" query. Seek and ye shall find.

English
1
0
2
165
pratyush รีทวีตแล้ว
Aritra
Aritra@ariG23498·
[HF ML Club India] We are pleased to announce the IRL event in Bengaluru. We are partnering with @GoogleDeepMind and this takes place on the 15th of July 2026 (more details below). We will have someone from the GDM team talk about Gemma 4, and an hour long build session with Gemma and @huggingface. Great opportunity for folks to network and build. PS: We will not entertain any "where is the registration link" query. Seek and ye shall find.
Aritra tweet media
English
32
12
306
23.2K