Z

3.7K posts

Z banner
Z

Z

@zdne

Founder/CTO @superfaceai – connecting AI to APIs so don't have to deal with other software anymore. Previously @good_api and @apiaryio.

Prague Katılım Ekim 2009
217 Takip Edilen1.4K Takipçiler
Z
Z@zdne·
@heyhey are you ok?
Z tweet media
English
0
0
0
47
Z
Z@zdne·
Enroute to NYC for the hackathon finale with @arc and @circle – can't wait to share more of what we are building!
Z tweet media
English
0
1
1
151
Z
Z@zdne·
@simonw I like the article @simonw but just to make sure - this is not MCP specific. Any agent that has access to tools is susceptible to prompt injection correct? As for the tool poisoning – that is addressed by using trusted tool providers (trusted MCP servers), correct?
English
1
0
1
118
Simon Willison
Simon Willison@simonw·
The GitHub MCP server suffers from the lethal trifecta for prompt injection: access to private data, exposure to malicious instructions and the ability to exfiltrate information. Be really careful with this stuff: attackers can trick your AI agent into stealing your private data
Luca Beurer-Kellner@lbeurerkellner

😈 BEWARE: Claude 4 + GitHub MCP will leak your private GitHub repositories, no questions asked. We discovered a new attack on agents using GitHub’s official MCP server, which can be exploited by attackers to access your private repositories. creds to @marco_milanta (1/n) 👇

English
13
82
433
57.2K
Z
Z@zdne·
Tomorrow, we will pack most of what I’ve learned about the AI use of APIs into ten points. Join us at maven.com/p/d4116d/ai-ag…
English
0
0
2
155
Z
Z@zdne·
Clever move from @Meta to add LLM into WhatsApp! Just let me know how to start a new chat and drop the memory!
Z tweet media
English
0
0
1
202
OpenTools
OpenTools@opentools_·
.@AnthropicAI will support stateless remote MCP servers with “just HTTP” as transport. We’re quite excited about this! So we gave a “lightning talk” at an MCP meetup hosted by @paulgb @CloudflareDev yesterday.
English
23
217
1.7K
100.8K
Z
Z@zdne·
Hello world from superface.ai agentic engine
English
0
0
1
135
Z
Z@zdne·
LLM function calling needs specially designed API connectors for AI agents to work well. Read more in this article: superface.ai/blog/ai-agent-…
English
0
0
0
131
Z
Z@zdne·
For example, when the function name is "microsoft-outlook__send-email__send-email" the function calling success rate is only around 50% if you change the function name to "microsoft-outlook__se__send-email" the success rate goes up to 70% but still nowehere near 4turbo
Z tweet media
English
0
0
0
121
Z
Z@zdne·
While most are praising GPT-4o my findings are that is a significant downgrade when it comes to function calling compared to 4turbo. If using external tools is something you need then make sure to check it twice before switching.
English
0
0
1
432
Z
Z@zdne·
🚀 We have launched the first API integration platform for AI! @superfaceai now provides unparalleled API connectivity for your LLM-based applications superface.ai/blog/hub-api Whether you need to use APIs in OpenAI, MistralAI, Anthropic, or LangChain, we've got your back ✅
English
0
1
2
567
Z
Z@zdne·
✔️Superface.ai now support function calling for @OpenAI, @MistralAI , @AnthropicAI , @cohere , and Google Gemini! This is the best and fastest way to connect all your models and the agents to APIs! Including solved end-user authentication of your agents.
English
0
0
0
180

Keşfet

@heyhey @arc @circle @simonw @Meta @superfaceai @opentools_ @AnthropicAI