Verwaltet von:
#errors
TC39 Advances “Array.fromAsync”, “Error.isError”, and Explicit Resource Management to Stage 4, by @sarahgooding (@SocketSecurity):
Designing type inference for high quality type errors
https://blog.polybdenum.com/2025/02/14/designing-type-inference-for-high-quality-type-errors.html
**Hallucinations are not the same as errors**
“_An AI can’t lie, because it doesn’t know it’s lying. In that sense, it’s sort of pathological, careening forward without a conscience._”
https://seths.blog/2025/05/hallucinations-are-not-the-same-as-errors/.
#AI #ArtificialIntelligence #Technology #Tech #Hallucinations #Errors #Lying #Lies @ai
Using tests as a debugging tool for logic errors
https://www.qodo.ai/blog/java-unit-testing-how-to-use-tests-as-a-debugging-tool-for-logic-errors/
A good example why #AI code generation introduces hard to recognize #errors and #bugs: https://www.bugsink.com/blog/copilot-induced-crash/
More and more stories about the thing that #LLM-generated code is even more difficult to debug because its results look so damn plausible but are wrong.
Therefore, it's really dangerous for non-trivial stuff.
How to Write Error Messages That Actually Help Users Rather Than Frustrate Them, by @Amy_Hupe (@piccalil.li):
Some of these apply beyond #Rust:
“Pitfalls Of Safe Rust”, Matthias Endler (https://corrode.dev/blog/pitfalls-of-safe-rust/).
Via HN: https://news.ycombinator.com/item?id=43585742
On Lobsters: https://lobste.rs/s/reznwf/pitfalls_safe_rust
Sometimes I’m playing with Kitten and something makes me smile. This time, it’s an error message. Always nice when you get a little hat-tip from past you to current you. Thanks for putting in the work, past me ;)
DOGE’s credibility is shot.
https://robertreich.substack.com/p/this-weeks-ten-reasons-for-modest
"DOGE has saved the govt just $2 billion. That comes to 1/35 of 1% of the federal budget.
Repeated #errors have raised questions about the quality & veracity of the #information Musk’s #DOGE is putting out. The mistakes also call into question the team members’ competence — whether they understand the #government well enough to cut it while avoiding catastrophe."
How to distrust a CA without any certificate errors — https://dadrian.io/blog/posts/sct-not-after/
#HackerNews #How #to #distrust #a #CA #without #any #certificate #errors #DistrustCA #CertificateErrors #CyberSecurity #HackerNews #SSL
→ When Your Last Name Is Null, Nothing Works
https://www.wsj.com/lifestyle/null-last-name-computer-scientists-forms-f0a43b08
“His last name is Null, the same word used by computer scientists to mean "no value" or "invalid value." The Nulls of the world, it turns out, endure a lifetime of website bouncebacks, processing errors and declarations by customer-service representatives that their accounts don’t exist.”
Falsehoods programmers believe about null pointers
https://purplesyringa.moe/blog/falsehoods-programmers-believe-about-null-pointers/
#Business #Pitfalls
AI mistakes differ sharply from human ones · Weirdness is what makes AI errors unique https://ilo.im/161y4o
_____
#Humans #Machines #Tools #AI #AiModels #Errors #Randomness #Trust #Confidence #Security
Study: Medical large language models are vulnerable to data-poisoning attacks
"replacement of just 0.001% of training tokens with medical #misinformation results in harmful models more likely to propagate medical #errors"
"a successful attack potentially executable for under US$1,000"
Bruce Schneier perfectly articulates something I’ve been thinking about:
AI Mistakes Are Very Different from Human Mistakes
https://www.schneier.com/blog/archives/2025/01/ai-mistakes-are-very-different-from-human-mistakes.html
This is exactly right, certainly for LLMs.
"It seems ridiculous when chatbots tell you to eat rocks or add glue to pizza. But it’s not the frequency or severity of AI systems’ mistakes that differentiates them from human mistakes. It’s their weirdness. AI systems do not make mistakes in the same ways that humans do.”
Required reading.
while i had heard that #Duolingo had introduced "AI" in some way, i had hoped it was just behind the scenes, and not able to inject #errors into the content.
"AI" in maths would be fatal indeed ...
I ve done two of the #language #courses and yes, there s errors there too, but those are older (and reporting them seems useless - never any reaction).
New blog post https://blog.cr.yp.to/20250118-flight.html "As expensive as a plane flight: Looking at some claims that quantum computers won't work." #quantum #energy #variables #errors #rsa #secrecy
