Trattow Pugliesi

739 posts

Trattow Pugliesi banner
Trattow Pugliesi

Trattow Pugliesi

@TraTTow_br

LLM Security Researcher Prompt injection, jailbreaks, agent abuse, and AI red teaming Breaking unsafe AI systems and studying how to defend them

Katılım Haziran 2023
745 Takip Edilen624 Takipçiler
Trattow Pugliesi
Trattow Pugliesi@TraTTow_br·
@aliByteCode Marketing is like a puzzle , it's about understanding people's needs, creating value, and connecting with your audience in a meaningful way. It's the art of sparking interest and building relationships through strategic communication.
GIF
English
0
0
1
9
Ali
Ali@aliByteCode·
what's your definition of marketing?
English
42
1
27
1.5K
Trattow Pugliesi
Trattow Pugliesi@TraTTow_br·
@Tobby_scraper Are you maximizing those deductions for your dev tools this tax season? It's like a refund waiting to happen!
English
0
0
0
9
Tobby_scraper
Tobby_scraper@Tobby_scraper·
GM Guys. Tax season. Are you expensing your dev tools?
English
3
0
5
736
Dmytro Chuta
Dmytro Chuta@dmitriychuta·
@TraTTow_br I don’t have a trial because I don’t sell a subscription, only a lifetime purchase 🙌
English
1
0
1
335
Dmytro Chuta
Dmytro Chuta@dmitriychuta·
✨I built a subscription tracker with the best import on the market, so users don’t have to enter everything manually. It’s fast and works really well, so I plan to let users try it before buying PRO in upcoming updates.
English
44
24
916
78.1K
Teodora @DesignerAnts
Teodora @DesignerAnts@designerants·
Some of my work from March! ✨ Doing lots of Apple Ads work lately.
English
11
10
156
25.4K
anul agarwal
anul agarwal@anulagarwal·
Pro tip for building mobile apps to increase conversion: Your first onboarding screen should always be a language selection screen. You don't want the user to follow the onboarding in a language they may not be comfortable in.
anul agarwal tweet media
English
68
8
298
47.8K
Trattow Pugliesi
Trattow Pugliesi@TraTTow_br·
@delveroin X is great for broader reach, while Reddit can offer more niche engagement. Consider your target audience and goals to make the best choice for distribution.
English
0
0
0
7
(Oma)devuae
(Oma)devuae@delveroin·
Let’s have a conversation: X or Reddit for distribution?
(Oma)devuae tweet media
English
111
1
109
8.6K
Trattow Pugliesi
Trattow Pugliesi@TraTTow_br·
@PaulSolt Saved here. I’ve been using Codex much more lately than Cursor, and with those skills it’ll fly. Thanks.
English
1
0
1
737
Trattow Pugliesi
Trattow Pugliesi@TraTTow_br·
@CaptainInsightX "Exfiltration" Your AI coding agent just ran 500 lines of code you didn't review. It also saved your AWS keys, API tokens, and that one variable named 'password=123' to a temp file it decided to 'optimize' into a commit. Welcome to vibe coding. 🙃
English
1
1
3
156
Captain Insight
Captain Insight@CaptainInsightX·
I’m a Vibe Coder, scare me with one word.
Captain Insight tweet media
English
99
4
124
3.8K
Bac Leo
Bac Leo@BacLeodiv·
Dear @X algo, I am looking to connect with people interested in: - Builders - AI Tech - Full Stack - Hackers - Building in Public Say hi & let's grow together 👋
English
124
0
89
5K
Eleftheria Batsou
Eleftheria Batsou@BatsouElef·
Share your product (free marketing). Support others in the comments. 🔗
English
337
0
145
13.8K
Trattow Pugliesi
Trattow Pugliesi@TraTTow_br·
This is exactly the pattern I keep seeing. OpenAI's bug bounty is a good start, but 50% reproducibility threshold? That's basically saying "we'll only fix it if it's easy to reproduce." The scary part about agents is they don't just fail once - they iterate. If there's a 1% bypass, they'll find it and run it 10,000 times.
English
0
0
0
75
Johann Rehberger
Johann Rehberger@wunderwuzzi23·
Sounds good, until you read the details. AI data exfil is now a "safety" issue, and only if the attack success rate is 50%+ Reminds me of early Windows ASLR with 256 options (<1% success rate). First real-world exploit bypassing it just ran in a loop. And, agents love loops.
OpenAI Newsroom@OpenAINewsroom

Today we’re launching a Safety Bug Bounty program focused on identifying AI abuse and safety risks across OpenAI products. This new program builds on our Security Bug Bounty to include AI-specific safety issues and misuse scenarios, helping us work with the safety and security research community to identify and address real-world risks. openai.com/index/safety-b…

English
4
4
32
6.7K
Trattow Pugliesi
Trattow Pugliesi@TraTTow_br·
Someone just named it: MaaS — Methodology as a Service CLAUDE.md isn't just coding rules. When you encode how an expert thinks — decision frameworks, evaluation criteria — you're shipping expertise as executable instructions. Here's how: No backend required Upload a CV → AI interview coach runs Structured methodology files + data folder Claude executes the expertise Domain experts (coaches, consultants, strategists) can ship their knowledge without writing code. The skill isn't programming. It's structuring expertise. Source: github.com/raphaotten/cla…
English
1
0
1
29
Trattow Pugliesi
Trattow Pugliesi@TraTTow_br·
Draft 1 - sift-kg (Knowledge Graphs from Docs) 🕸️ Your docs → interactive knowledge graph Built sift-kg - CLI tool that: • Extracts entities and relations from document collections • Builds a browsable knowledge graph • Works with ANY LLM via LiteLLM (including Ollama) • Your docs never leave your machine Here's how: pip install sift-kg sift extract ./docs/ --model ollama/llama3 sift build && sift view Tested: 9 FTX articles → 373 entities, 1,184 relations. Finally, something that makes sense of all those docs. Source: github.com/juanceresa/sif…
English
0
1
0
51