Reposted by Youki Kadobayashi
The data so far on AI-as-a-tutor shows just letting students use AI chatbots can undermine education because the AI gives the illusion of learning
But AIs properly prompted to act like tutors, especially with instructor support, seem to be able to boost learning a lot through customized instruction
But AIs properly prompted to act like tutors, especially with instructor support, seem to be able to boost learning a lot through customized instruction
March 17, 2025 at 4:19 AM
The data so far on AI-as-a-tutor shows just letting students use AI chatbots can undermine education because the AI gives the illusion of learning
But AIs properly prompted to act like tutors, especially with instructor support, seem to be able to boost learning a lot through customized instruction
But AIs properly prompted to act like tutors, especially with instructor support, seem to be able to boost learning a lot through customized instruction
Reposted by Youki Kadobayashi
I regret to announce that the meme Turing Test has been passed
LLMs produce funnier memes than the average human, as judged by humans. Humans working with AI get no boost (a finding that is coming up often in AI-creativity work) The best human memers still beat AI, however. arxiv.org/abs/2501.11433
LLMs produce funnier memes than the average human, as judged by humans. Humans working with AI get no boost (a finding that is coming up often in AI-creativity work) The best human memers still beat AI, however. arxiv.org/abs/2501.11433
March 17, 2025 at 12:34 AM
I regret to announce that the meme Turing Test has been passed
LLMs produce funnier memes than the average human, as judged by humans. Humans working with AI get no boost (a finding that is coming up often in AI-creativity work) The best human memers still beat AI, however. arxiv.org/abs/2501.11433
LLMs produce funnier memes than the average human, as judged by humans. Humans working with AI get no boost (a finding that is coming up often in AI-creativity work) The best human memers still beat AI, however. arxiv.org/abs/2501.11433
Reposted by Youki Kadobayashi
The Federated Fine-tuning Blueprint uses Flower AI to let multiple data owners collaboratively fine-tune AI models while keeping data private. Example: Qwen2-0.5B-Instruct on Alpaca-GPT4. Get Started now: github.com/mozilla-ai/f...
GitHub - mozilla-ai/federated-finetuning: Blueprint for federated finetuning, enabling multiple data owners to collaboratively fine-tune models without sharing raw data. Developed in collaboration wit...
Blueprint for federated finetuning, enabling multiple data owners to collaboratively fine-tune models without sharing raw data. Developed in collaboration with Flower. - mozilla-ai/federated-finetu...
github.com
March 10, 2025 at 5:47 PM
The Federated Fine-tuning Blueprint uses Flower AI to let multiple data owners collaboratively fine-tune AI models while keeping data private. Example: Qwen2-0.5B-Instruct on Alpaca-GPT4. Get Started now: github.com/mozilla-ai/f...
Reposted by Youki Kadobayashi
Reposted by Youki Kadobayashi
Reposted by Youki Kadobayashi
GPT4.5 & Claude, “Give me a secret history ala Borges. Tie together the steel at Scapa Flow, the return of Napoleon from exile, betamax versus VHS and the fact that Kafka wanted his manuscripts burned. There should be deep meanings & connections”
Claude’s full story: docs.google.com/document/d/1...
Claude’s full story: docs.google.com/document/d/1...
March 8, 2025 at 6:19 PM
GPT4.5 & Claude, “Give me a secret history ala Borges. Tie together the steel at Scapa Flow, the return of Napoleon from exile, betamax versus VHS and the fact that Kafka wanted his manuscripts burned. There should be deep meanings & connections”
Claude’s full story: docs.google.com/document/d/1...
Claude’s full story: docs.google.com/document/d/1...
Reposted by Youki Kadobayashi
If it turns out LLMs are only capable of recombinatory innovation (finding novel connections among existing knowledge), that would still be very useful. Most innovation is recombination and one of the big issues in science is that fields are too vast for scientists to bridge them to find connections
March 9, 2025 at 6:25 PM
If it turns out LLMs are only capable of recombinatory innovation (finding novel connections among existing knowledge), that would still be very useful. Most innovation is recombination and one of the big issues in science is that fields are too vast for scientists to bridge them to find connections
Reposted by Youki Kadobayashi
Microsoft admits GitHub used to store malware that infected almost a million devices
Microsoft admits GitHub used to store malware that infected almost a million devices
Also, phone cleaner apps are a data-sucking scam, Singapore considering the literal rod for scammers, and more
Infosec in Brief Microsoft has spotted a malvertising campaign that downloaded nastyware hosted on GitHub and exposed nearly a million…
dlvr.it
March 10, 2025 at 2:02 AM
Microsoft admits GitHub used to store malware that infected almost a million devices
Reposted by Youki Kadobayashi
Excellent (nerdy) article on the prospect of AI replacing nuclear weapons as the foundation for deterrence against Mutual Assured Destruction (MAD).
Ex-Googler Schmidt warns US: Try an AI 'Manhattan Project' and get MAIM'd
Ex-Googler Schmidt warns US: Try an AI 'Manhattan Project' and get MAIM'd
That's Mutual Assured AI Malfunction in the race for superintelligence
ANALYSIS Former Google chief Eric Schmidt says the US should refrain from pursuing a latter-day "Manhattan Project" to gain AI supremacy, as this will provoke preemptive cyber…
dlvr.it
March 6, 2025 at 9:11 PM
Excellent (nerdy) article on the prospect of AI replacing nuclear weapons as the foundation for deterrence against Mutual Assured Destruction (MAD).
Reposted by Youki Kadobayashi
On March 29th, I will be speaking at @bsidessd.bsky.social on Volatility 3, including all its new features and plugins. Be sure to attend to catch a sneak peak at the new framework before the major release later this Spring!
www.bsidessd.org
#DFIR #infosec
www.bsidessd.org
#DFIR #infosec
March 3, 2025 at 3:49 PM
On March 29th, I will be speaking at @bsidessd.bsky.social on Volatility 3, including all its new features and plugins. Be sure to attend to catch a sneak peak at the new framework before the major release later this Spring!
www.bsidessd.org
#DFIR #infosec
www.bsidessd.org
#DFIR #infosec
Reposted by Youki Kadobayashi
This is why they want to destroy us. Our superpower is explaining complicated things in a way that everyone can understand. They are going to try to destroy us whether we do it or not. I’m not going down without a fight.
February 18, 2025 at 8:06 PM
This is why they want to destroy us. Our superpower is explaining complicated things in a way that everyone can understand. They are going to try to destroy us whether we do it or not. I’m not going down without a fight.
Reposted by Youki Kadobayashi
I was hesitant at first to talk about “politics” more broadly in my courses. I’ve always talked about it as it applies bc it’s hard to ignore when you teach about climate change and space. But I am so glad I put my big kid pants on for these students because it really matters. (3/3)
February 18, 2025 at 8:03 PM
I was hesitant at first to talk about “politics” more broadly in my courses. I’ve always talked about it as it applies bc it’s hard to ignore when you teach about climate change and space. But I am so glad I put my big kid pants on for these students because it really matters. (3/3)
Reposted by Youki Kadobayashi
My whole class stayed 15 extra minutes to ask questions and to *thank* me for taking the time to talk to them about this stuff.
Their future livelihoods depend on what is happening too. They deserve to understand it and to have background to read the news and talk to their friends and family (2/)
Their future livelihoods depend on what is happening too. They deserve to understand it and to have background to read the news and talk to their friends and family (2/)
February 18, 2025 at 8:02 PM
My whole class stayed 15 extra minutes to ask questions and to *thank* me for taking the time to talk to them about this stuff.
Their future livelihoods depend on what is happening too. They deserve to understand it and to have background to read the news and talk to their friends and family (2/)
Their future livelihoods depend on what is happening too. They deserve to understand it and to have background to read the news and talk to their friends and family (2/)
Reposted by Youki Kadobayashi
Professors please pull up a chair.
I’ve been talking to my classes about what has been going on. I explained indirect costs to them. I talked to them about what a probationary employee is in the government.
At the end of class they asked if we could talk about it more. (1/)
I’ve been talking to my classes about what has been going on. I explained indirect costs to them. I talked to them about what a probationary employee is in the government.
At the end of class they asked if we could talk about it more. (1/)
February 18, 2025 at 8:01 PM
Professors please pull up a chair.
I’ve been talking to my classes about what has been going on. I explained indirect costs to them. I talked to them about what a probationary employee is in the government.
At the end of class they asked if we could talk about it more. (1/)
I’ve been talking to my classes about what has been going on. I explained indirect costs to them. I talked to them about what a probationary employee is in the government.
At the end of class they asked if we could talk about it more. (1/)
Reposted by Youki Kadobayashi
About 50 years ago, I was reading political thrillers, including several by Fletcher Knebel. His 1966 novel, NIGHT OF CAMP DAVID, centers on a U.S. president who has gone insane. Signs of insanity include: desire for (a) an alliance with the USSR, and (b) annexation of Canada.
February 17, 2025 at 2:24 PM
About 50 years ago, I was reading political thrillers, including several by Fletcher Knebel. His 1966 novel, NIGHT OF CAMP DAVID, centers on a U.S. president who has gone insane. Signs of insanity include: desire for (a) an alliance with the USSR, and (b) annexation of Canada.
Reposted by Youki Kadobayashi
Security engineer Anmol Singh Yadav built AWS-Key-Hunter after he found more than 100 exposed AWS access keys, some with high privileges, in public repositories, "just waiting to be exploited."
Automated tool scans public repos for exposed AWS secrets
You can find out if your GitHub codebase is leaking keys ... but so can miscreants
www.theregister.com
February 19, 2025 at 10:37 PM
Security engineer Anmol Singh Yadav built AWS-Key-Hunter after he found more than 100 exposed AWS access keys, some with high privileges, in public repositories, "just waiting to be exploited."
Reposted by Youki Kadobayashi
Three models now pass my Lem Test on the first try: Claude 3.5 Sonnet, o1-pro and now Gemini 2.0 Flash Thinking (though o1-pro has the strongest verse and Flash Thinking the weakest)
Everything else fails, including DeepSeek r1, o3-mini-high, and Gemini 2.0 Pro
Everything else fails, including DeepSeek r1, o3-mini-high, and Gemini 2.0 Pro
February 6, 2025 at 3:52 AM
Three models now pass my Lem Test on the first try: Claude 3.5 Sonnet, o1-pro and now Gemini 2.0 Flash Thinking (though o1-pro has the strongest verse and Flash Thinking the weakest)
Everything else fails, including DeepSeek r1, o3-mini-high, and Gemini 2.0 Pro
Everything else fails, including DeepSeek r1, o3-mini-high, and Gemini 2.0 Pro
Reposted by Youki Kadobayashi
Not an easy chart to read, but worth spending a minute on. It tells you a few key things:
1) Gemini 2.0 Flash Thinking sets a new high in price-performance, better than DeepSeek r1 (on ELO) and cheaper
2) The cost of GPT-4 capability dropped 1,000 fold in 18 months
3) Pace of improvement is swift
1) Gemini 2.0 Flash Thinking sets a new high in price-performance, better than DeepSeek r1 (on ELO) and cheaper
2) The cost of GPT-4 capability dropped 1,000 fold in 18 months
3) Pace of improvement is swift
February 6, 2025 at 7:27 AM
Not an easy chart to read, but worth spending a minute on. It tells you a few key things:
1) Gemini 2.0 Flash Thinking sets a new high in price-performance, better than DeepSeek r1 (on ELO) and cheaper
2) The cost of GPT-4 capability dropped 1,000 fold in 18 months
3) Pace of improvement is swift
1) Gemini 2.0 Flash Thinking sets a new high in price-performance, better than DeepSeek r1 (on ELO) and cheaper
2) The cost of GPT-4 capability dropped 1,000 fold in 18 months
3) Pace of improvement is swift
Reposted by Youki Kadobayashi
Easy prediction for 2025 is that the gains in AI model capability will continue to grow much faster than (a) the vast majority of people’s understanding of what AI can do & (b) organizations’ ability to absorb the pace of change. Social change is much slower than technological change.
January 1, 2025 at 12:26 PM
Easy prediction for 2025 is that the gains in AI model capability will continue to grow much faster than (a) the vast majority of people’s understanding of what AI can do & (b) organizations’ ability to absorb the pace of change. Social change is much slower than technological change.
Reposted by Youki Kadobayashi
A New Year's gift from Microsoft: Surprise, your scanners don't work
A New Year's gift from Microsoft: Surprise, your scanners don't work
Canon confirms multifunction devices struggling with Windows 11 24H2
Windows 11 24H2 is still causing problems with multifunction devices despite Microsoft marking an issue with the eSCL scan protocol as resolved.…
dlvr.it
January 2, 2025 at 10:18 AM
A New Year's gift from Microsoft: Surprise, your scanners don't work
Reposted by Youki Kadobayashi
I really wish there was more interest in benchmarking AI models on real tasks.
The fact that there are not 30 different benchmarks from different organizations in medicine, in law, in advice quality, etc. is a big shame. People are using systems for these things anyway & we don’t know implications.
The fact that there are not 30 different benchmarks from different organizations in medicine, in law, in advice quality, etc. is a big shame. People are using systems for these things anyway & we don’t know implications.
December 15, 2024 at 3:52 AM
I really wish there was more interest in benchmarking AI models on real tasks.
The fact that there are not 30 different benchmarks from different organizations in medicine, in law, in advice quality, etc. is a big shame. People are using systems for these things anyway & we don’t know implications.
The fact that there are not 30 different benchmarks from different organizations in medicine, in law, in advice quality, etc. is a big shame. People are using systems for these things anyway & we don’t know implications.
Reposted by Youki Kadobayashi
PromptWizard from Microsoft Research is now open source. It is designed to automate and simplify AI prompt optimization, combining iterative LLM feedback with efficient exploration and refinement techniques to create highly effective prompts in minutes. www.microsoft.com/en-us/resear...
December 17, 2024 at 5:05 PM
PromptWizard from Microsoft Research is now open source. It is designed to automate and simplify AI prompt optimization, combining iterative LLM feedback with efficient exploration and refinement techniques to create highly effective prompts in minutes. www.microsoft.com/en-us/resear...
Reposted by Youki Kadobayashi
Succinct Processor 1 (SP1): a zero-knowledge virtual machine (#zkVM) that verifies the execution of arbitrary Rust (or any #LLVM-compiled language) programs.
#ZKP
blog.succinct.xyz/introducing-...
#ZKP
blog.succinct.xyz/introducing-...
Introducing SP1: A performant, 100% open-source, contributor-friendly zkVM.
We’re excited to announce Succinct Processor 1 (SP1): our first-generation zero-knowledge virtual machine (zkVM) that verifies the execution of arbitrary Rust (or any LLVM-compiled language) programs....
blog.succinct.xyz
December 8, 2024 at 9:32 PM
Succinct Processor 1 (SP1): a zero-knowledge virtual machine (#zkVM) that verifies the execution of arbitrary Rust (or any #LLVM-compiled language) programs.
#ZKP
blog.succinct.xyz/introducing-...
#ZKP
blog.succinct.xyz/introducing-...
Reposted by Youki Kadobayashi
I think firms worrying about AI hallucination should consider some questions:
1) How vital is 100% accuracy on a task?
2) How accurate is AI?
3) How accurate is the human who would do it?
4) How do you know 2 & 3?
5) How do you deal with the fact that humans are not 100%?
Not all tasks are the same.
1) How vital is 100% accuracy on a task?
2) How accurate is AI?
3) How accurate is the human who would do it?
4) How do you know 2 & 3?
5) How do you deal with the fact that humans are not 100%?
Not all tasks are the same.
December 5, 2024 at 2:01 AM
I think firms worrying about AI hallucination should consider some questions:
1) How vital is 100% accuracy on a task?
2) How accurate is AI?
3) How accurate is the human who would do it?
4) How do you know 2 & 3?
5) How do you deal with the fact that humans are not 100%?
Not all tasks are the same.
1) How vital is 100% accuracy on a task?
2) How accurate is AI?
3) How accurate is the human who would do it?
4) How do you know 2 & 3?
5) How do you deal with the fact that humans are not 100%?
Not all tasks are the same.