Heidi
@heidisov.bsky.social
Here for the ML/AI discussions. #Bitcoin and human rights.
Reposted by Heidi
A petition for Francesca Albanese and Gaza doctors to win a Nobel Peace Prize has gained over one million signatures
Over one million back Nobel Peace Prize for Francesca Albanese and Gaza doctors
www.thenational.scot
September 25, 2025 at 12:23 PM
A petition for Francesca Albanese and Gaza doctors to win a Nobel Peace Prize has gained over one million signatures
Reposted by Heidi
I did an AMA on Stacker News! 🧡⚡
I shared insights from 5 months in Southern Africa - diving into Bitcoin adoption, remittances, self-custody challenges, and grassroots impact. 🌍
💬 Read it here →
I shared insights from 5 months in Southern Africa - diving into Bitcoin adoption, remittances, self-custody challenges, and grassroots impact. 🌍
💬 Read it here →
Hi, it's Anita Posch! After 5 months in southern Africa AMA about adoption & BTC \ stacker news ~AMA
Happy to be here! [58 comments]
stacker.news
June 17, 2025 at 9:02 AM
I did an AMA on Stacker News! 🧡⚡
I shared insights from 5 months in Southern Africa - diving into Bitcoin adoption, remittances, self-custody challenges, and grassroots impact. 🌍
💬 Read it here →
I shared insights from 5 months in Southern Africa - diving into Bitcoin adoption, remittances, self-custody challenges, and grassroots impact. 🌍
💬 Read it here →
The media also did/does this…
LLMs are getting worse at summarizing scientific data: “Generalization bias in large language model summarization of scientific research”
doi.org/10.1098/rsos...
doi.org/10.1098/rsos...
Generalization bias in large language model summarization of scientific research | Royal Society Open Science
Artificial intelligence chatbots driven by large language models (LLMs) have the potential
to increase public science literacy and support scientific research, as they can quickly
summarize complex sc...
doi.org
May 19, 2025 at 12:14 PM
The media also did/does this…
People don’t seem to get that when an LLM finds that error in the code for you (and quicker than you ever could), they are in fact correcting YOUR incorrect information.
May 18, 2025 at 8:17 PM
People don’t seem to get that when an LLM finds that error in the code for you (and quicker than you ever could), they are in fact correcting YOUR incorrect information.
Reposted by Heidi
MIT just dropped a clever new way to make AI write better code 🤖💻. Big win for efficiency 👏 💡💻 #AI #LLM #MIT #MachineLearning
news.mit.edu/2025/making-...
news.mit.edu/2025/making-...
Making AI-generated code more accurate in any language
Researchers developed a more efficient way to control the outputs of a large language model, guiding it to generate text that adheres to a certain structure, like a programming language, and remains e...
news.mit.edu
April 21, 2025 at 6:44 PM
MIT just dropped a clever new way to make AI write better code 🤖💻. Big win for efficiency 👏 💡💻 #AI #LLM #MIT #MachineLearning
news.mit.edu/2025/making-...
news.mit.edu/2025/making-...
Reposted by Heidi
Really good analysis about common pitfalls LLM AI generated code and their impact. While these tools help developers being more productive, do not blindly trust the provided code as it might not be the best in multiple levels.
martinfowler.com/articles/exp...
martinfowler.com/articles/exp...
The role of developer skills in agentic coding
Notes from my Thoughtworks colleagues on AI-assisted software delivery
martinfowler.com
April 19, 2025 at 4:21 PM
Really good analysis about common pitfalls LLM AI generated code and their impact. While these tools help developers being more productive, do not blindly trust the provided code as it might not be the best in multiple levels.
martinfowler.com/articles/exp...
martinfowler.com/articles/exp...
Reposted by Heidi
Mitigating Security Risks in MCP Implementations
MCP (Microsoft Cloud Platform) provides a robust framework for integrating LLM applications with various tools and data sources. However, improper authorization logic within the MCP server may result in sensitive data exposure and misapplied access…
MCP (Microsoft Cloud Platform) provides a robust framework for integrating LLM applications with various tools and data sources. However, improper authorization logic within the MCP server may result in sensitive data exposure and misapplied access…
Mitigating Security Risks in MCP Implementations
MCP (Microsoft Cloud Platform) provides a robust framework for integrating LLM applications with various tools and data sources. However, improper authorization logic within the MCP server may result in sensitive data exposure and misapplied access controls. To uphold security and compliance, the following KQL query helps detect newly deployed Azure MCP instances within your MDE (Microsoft Defender for Endpoint) environment.
undercodetesting.com
April 20, 2025 at 1:26 PM
Mitigating Security Risks in MCP Implementations
MCP (Microsoft Cloud Platform) provides a robust framework for integrating LLM applications with various tools and data sources. However, improper authorization logic within the MCP server may result in sensitive data exposure and misapplied access…
MCP (Microsoft Cloud Platform) provides a robust framework for integrating LLM applications with various tools and data sources. However, improper authorization logic within the MCP server may result in sensitive data exposure and misapplied access…
Reposted by Heidi
1. LLM-generated code tries to run code from online software packages. Which is normal but
2. The packages don’t exist. Which would normally cause an error but
3. Nefarious people have made malware under the package names that LLMs make up most often. So
4. Now the LLM code points to malware.
2. The packages don’t exist. Which would normally cause an error but
3. Nefarious people have made malware under the package names that LLMs make up most often. So
4. Now the LLM code points to malware.
LLMs hallucinating nonexistent software packages with plausible names leads to a new malware vulnerability: "slopsquatting."
LLMs can't stop making up software dependencies and sabotaging everything
: Hallucinated package names fuel 'slopsquatting'
www.theregister.com
April 12, 2025 at 11:43 PM
1. LLM-generated code tries to run code from online software packages. Which is normal but
2. The packages don’t exist. Which would normally cause an error but
3. Nefarious people have made malware under the package names that LLMs make up most often. So
4. Now the LLM code points to malware.
2. The packages don’t exist. Which would normally cause an error but
3. Nefarious people have made malware under the package names that LLMs make up most often. So
4. Now the LLM code points to malware.
Reposted by Heidi
Stanford researchers developed a robust steganography system leveraging language models, embedding hidden messages in text. This method resists common attacks like paraphrasing, enhancing security in environments where encryption can attract unwanted attention. https://arxiv.org/abs/2504.08977
Robust Steganography from Large Language Models
ArXiv link for Robust Steganography from Large Language Models
arxiv.org
April 15, 2025 at 11:50 PM
Stanford researchers developed a robust steganography system leveraging language models, embedding hidden messages in text. This method resists common attacks like paraphrasing, enhancing security in environments where encryption can attract unwanted attention. https://arxiv.org/abs/2504.08977
Reposted by Heidi
Finally found the time to watch it in full: one of the most interesting and thought provoking LLM conference I’ve seen in a while. You see LLM as latent attention graphs and many structural oddities suddenly falls into place. www.youtube.com/watch?v=J1YC...
April 12, 2025 at 6:45 AM
Finally found the time to watch it in full: one of the most interesting and thought provoking LLM conference I’ve seen in a while. You see LLM as latent attention graphs and many structural oddities suddenly falls into place. www.youtube.com/watch?v=J1YC...
Reposted by Heidi
MCP Security Attacks
invariantlabs.ai/blog/mcp-sec...
invariantlabs.ai/blog/mcp-sec...
April 6, 2025 at 5:48 PM
MCP Security Attacks
invariantlabs.ai/blog/mcp-sec...
invariantlabs.ai/blog/mcp-sec...
Dropping mid-day on a Saturday when many people are “hands off”
🚨Llama 4 Is Out!🚨
2 out of 3 models just released
- Scout: 109B / 17B active
- Maverick: 400B / 17B active
- Bohemoth: 2T / 288B active
ai.meta.com/blog/llama-4...
2 out of 3 models just released
- Scout: 109B / 17B active
- Maverick: 400B / 17B active
- Bohemoth: 2T / 288B active
ai.meta.com/blog/llama-4...
The Llama 4 herd: The beginning of a new era of natively multimodal AI innovation
We’re introducing Llama 4 Scout and Llama 4 Maverick, the first open-weight natively multimodal models with unprecedented context support and our first built using a mixture-of-experts (MoE) architect...
ai.meta.com
April 5, 2025 at 11:52 PM
Dropping mid-day on a Saturday when many people are “hands off”
Reposted by Heidi
It’s so silly that tech people keep saying that we’re on the verge of a full artificial intelligence takeover and yet most automatic soap dispensers can’t tell hand from not hand half the time.
April 4, 2025 at 1:51 PM
It’s so silly that tech people keep saying that we’re on the verge of a full artificial intelligence takeover and yet most automatic soap dispensers can’t tell hand from not hand half the time.
Reposted by Heidi
400 videos and counting! 🎥🔥 Let’s take it to 4,000 subscribers!
If you love #Bitcoin and financial freedom, hit subscribe and join the journey! ⚡💡 www.youtube.com/anitaposch
If you love #Bitcoin and financial freedom, hit subscribe and join the journey! ⚡💡 www.youtube.com/anitaposch
March 25, 2025 at 9:00 AM
400 videos and counting! 🎥🔥 Let’s take it to 4,000 subscribers!
If you love #Bitcoin and financial freedom, hit subscribe and join the journey! ⚡💡 www.youtube.com/anitaposch
If you love #Bitcoin and financial freedom, hit subscribe and join the journey! ⚡💡 www.youtube.com/anitaposch
Friendly reminder to exclude your locally downloaded models folder from your backup process.
Just eats up space.
Just eats up space.
March 29, 2025 at 4:48 PM
Friendly reminder to exclude your locally downloaded models folder from your backup process.
Just eats up space.
Just eats up space.
Reposted by Heidi
A beginner’s guide to using Signal
A beginner’s guide to using Signal
You don’t have to be a government official to use it.
buff.ly
March 28, 2025 at 5:10 PM
A beginner’s guide to using Signal
At this rate, government will soon require ML/AI be wiped of all training data that is DEI.
We will be back to the old NLP processing of:
white man is to doctor as white woman is to nurse.
We will be back to the old NLP processing of:
white man is to doctor as white woman is to nurse.
March 29, 2025 at 12:07 AM
At this rate, government will soon require ML/AI be wiped of all training data that is DEI.
We will be back to the old NLP processing of:
white man is to doctor as white woman is to nurse.
We will be back to the old NLP processing of:
white man is to doctor as white woman is to nurse.
Interesting list, regardless of your perspectives: climatejets.org
Private Jet Emissions - ClimateJets
Ever wondered about the carbon emissions of the ultra-rich? Explore the climate implications of private aircraft at ClimateJets.
climatejets.org
March 4, 2025 at 1:08 PM
Interesting list, regardless of your perspectives: climatejets.org
Reposted by Heidi
Fantastic thread
Musk has no technical skills whatsoever, but he wants to appear smart. So he takes bits of information like this, told to him by junior engineers, and regurgitates it to appear smart.
Musk did this with the Twitter stack and Twitter's senior architects called him out publicly, then he fired them.
Musk did this with the Twitter stack and Twitter's senior architects called him out publicly, then he fired them.
One of Musk's defining qualities is that he is a conspiracy theorist when it comes to government. He will see some ambiguous information or be presented with an easily debunked tale and he immediately believes the worst.
February 9, 2025 at 10:23 PM
Fantastic thread
Reposted by Heidi
Finally got this prototype to work after 1000 years of debugging. Sweet release 🥲
The idea is I dump a bunch of unstructured notes on a topic in, feed that into gpt-4o-mini, ask it to label sections with a set of types – claim, evidence, assumption, etc. – and display them as coloured highlights
The idea is I dump a bunch of unstructured notes on a topic in, feed that into gpt-4o-mini, ask it to label sections with a set of types – claim, evidence, assumption, etc. – and display them as coloured highlights
January 31, 2025 at 11:27 AM
Finally got this prototype to work after 1000 years of debugging. Sweet release 🥲
The idea is I dump a bunch of unstructured notes on a topic in, feed that into gpt-4o-mini, ask it to label sections with a set of types – claim, evidence, assumption, etc. – and display them as coloured highlights
The idea is I dump a bunch of unstructured notes on a topic in, feed that into gpt-4o-mini, ask it to label sections with a set of types – claim, evidence, assumption, etc. – and display them as coloured highlights
Reposted by Heidi
Let's make AI more inclusive.
At @huggingface.bsky.social we'll launch a huge community sprint soon to build high-quality training datasets for many languages.
We're looking for Language Leads to help with outreach.
Find your language and nominate yourself:
forms.gle/iAJVauUQ3FN8...
At @huggingface.bsky.social we'll launch a huge community sprint soon to build high-quality training datasets for many languages.
We're looking for Language Leads to help with outreach.
Find your language and nominate yourself:
forms.gle/iAJVauUQ3FN8...
November 26, 2024 at 6:29 AM
Let's make AI more inclusive.
At @huggingface.bsky.social we'll launch a huge community sprint soon to build high-quality training datasets for many languages.
We're looking for Language Leads to help with outreach.
Find your language and nominate yourself:
forms.gle/iAJVauUQ3FN8...
At @huggingface.bsky.social we'll launch a huge community sprint soon to build high-quality training datasets for many languages.
We're looking for Language Leads to help with outreach.
Find your language and nominate yourself:
forms.gle/iAJVauUQ3FN8...