bildung.social ist einer von vielen unabhängigen Mastodon-Servern, mit dem du dich im Fediverse beteiligen kannst.
Bildung unter den Bedingungen der digitalen Transformation.

Verwaltet von:

Serverstatistik:

856
aktive Profile

#claudeai

1 Beitrag1 Beteiligte*r0 Beiträge heute

Anthropic оновила свою платформу для розробників, додавши до неї нові функції командної співпраці, зокрема розширивши можливості міркувань для свого асистента Claude AI, долаючи ключові виклики у впровадженні штучного інтелекту на підприємствах.

#AI #Anthropic #ClaudeAI #LLM #ШІ

thetransmitted.com/ai/anthropi

Antwortete im Thread

@carnage4life I don't know what motivated him to say this. 😏

None of the coding is at this level, decades off being used on critical apps.

It NEVER works 1st time, takes LONGER to fix, writes side-hobby level code, frequently hits valleys it cannot get out of, often ignores what you tell it, and has very limited context.

It answers the question it can answer fastest closest to the question you asked, not the question you asked.

Pure, unadulterated bullshit.

Remember that excellent (11-year-old!) comedy sketch "The Expert"? I'm probably not the first to think of this, but I decided to see how an #LLM would handle the (kinda-)impossible challenge.

The answer is: pretty much how you'd expect. Because they're eager to provide an answer, even where an answer doesn't exist, many #AI will give an invalid answer rather than no answer.

#ClaudeAI was an exception, but that doesn't change the moral of the story: you can't trust #GenAI to do something for you that you couldn't do for yourself.

🔗 Much more, plus screenshots and podcast variant: danq.me/ai-vs-the-expert/

Dan Q · AI vs The ExpertInspired by an 11-year old comedy sketch, I asked a GenAI to solve an unsolvable programming problem... and (for at least some models) it failed in exactly the way I anticipated: claiming to be able to solve it and delivering code that just... didn't. What does this teach us about AI trustworthiness for problems that might be solvable, but for which the human operator doesn't have sufficient comprehension to verify?

#goodnews (thx #claude #claudeai #ai)

Typical situations where the Logistic Map (logistic function, s-function, sigmoid, expit) is used are:
- epidemics
- (bacterial) growth
- product life cycles
- language acquisition
- fox-rabbit-simulations
- many more en.wikipedia.org/wiki/Logistic

The curve over time grows almost exponentially when there is plenty of resources and falls flat when resources depleat.

I asked Claude to fit such a curve on global energy consumption. Claude answered with a Python script. I tweaked the x range and got the chart shown.

EDIT: The historic numbers weren't the aggregated sum, so we actually have to fit the sigmoid derivative.

Anthropic Challenges Hackers with $15,000 Bounty to Jailbreak New AI Safety System

Anthropic has unveiled a groundbreaking AI safety system that utilizes Constitutional Classifiers to thwart jailbreak attempts on its Claude model. With a $15,000 reward on the line, the company is in...

news.lavx.hu/article/anthropic

#news#tech#ConstitutionalAI

Dopo qualche mese di utilizzo di #claudeai ho deciso di provare #chatgpt
Sono passati pochi giorni ma già noto alcune differenze:
- Claude interpreta correttamente i miei prompt (o almeno quello che intendo) rispetto a chatgpt
- Claude fornisce soluzioni con codice subito al primo prompt e sono abbastanza adeguate al primo colpo. Chatgpt deve essere pregato o insultato per scrivere 4 righe di script
- ChatGPT (4/o1) è uno sbolognatore seriale di lavoro e non da soluzioni immediate.
Vi Torna?