carl24k
@carl24k.bsky.social
150 followers 280 following 290 posts
#DataScience & #MachineLearning - #Neuroscience & #NeuroAI - Author of Fighting Churn With Data - https://linktr.ee/carl24k
Posts Media Videos Starter Packs
carl24k.bsky.social
Is there an organization you would recommend for foreigners to make donations to help?
carl24k.bsky.social
More news from the intersection of AI and churn, my two favorite subjects: Vibe coding usage plunged after summer high.

- Lovable down 40% since its peak
- Vercel's v0 visits plunge 64%
- Bolt.new slipped 27% since June

www.businessinsider.com/ai-vibe-codi...

#AI #genai #churn #aihype
carl24k.bsky.social
Glad you’re okay!
carl24k.bsky.social
and I'm like "why am I paying for this when the free version is enough?" I guess I can keep this up until enshitification sets in and they start showing ads! 😂 What does your GenAI subscriber behavior look like?
carl24k.bsky.social
I am officially a serial churner from GenAI products - in the last 6 months I have signed up and then canceled a month each of both Claude and Gemini. I was working on specific projects, hit the limits, and I signed up. Then I finished the projects, my usage dropped...

#genai #churn
carl24k.bsky.social
"AI seems to have reached its iPhone 4 moment", meaning noticeable slowdown in improvements (for those of you old enough to remember the iPhone 4 ;-)

".this sort of feeling of disappointment is due to unreasonable levels of hype." The culprits? Usual suspects: Altman, Musk, Amodei (Anthropic)
carl24k.bsky.social
3. Neusosymbolic AI - combine statistical approaches with abstract reasoning logic, like humans do

Billions have already been wasted! But will investors learn and stop throwing good money after bad?
carl24k.bsky.social
Marcus recommends three new approaches to invest in, all based on scientific evidence from human cognition:

1. AI that builds World models a relationship between concepts the way humans do

2. Building in basic concepts like time, space and causality, like human infants have
carl24k.bsky.social
That scaling has plateaued is clear from the GPT-5 and llama fiasco‘s. Billions were spent for very minimal improvements. Marcus does engage in a little “I told you so” victory lap here referring to his 2022 assertions that were ridiculed back then. Can’t really blame him - Who’s laughing now?
carl24k.bsky.social
Are you an LLM AGI skeptic? Leading LLM AGI skeptic Gary Marcus has posted an op Ed in the New York Times:

TLDR : scaling has really run out of steam. New approaches are needed.

www.nytimes.com/2025/09/03/o...

#aihype #llm #agi
Opinion | The Fever Dream of Imminent Superintelligence Is Finally Breaking
www.nytimes.com
carl24k.bsky.social
business owners and customers are humans and they shouldn’t have to bear the cost. I live in Berkeley so trust me I know what I’m talking about. You probably live in a gated community You Bastard 😉
carl24k.bsky.social
I’m going to stick my neck out here: the camps do real economic harm to nearby businesses. The reality is people don’t want to walk past camps to shop or rent an apartment. If you want to, invite some unhoused to camp at your place. But it needs to be banned and enforced anywhere near businesses.
carl24k.bsky.social
Need some last minute reading for your Labor Day Weekend? Just in time, its @manning.com #labordaysale! Labor can meet the beach as you #learntocode or #learndatascience this weekend! www.manning.com/books/fighti...
carl24k.bsky.social
Are #AI and #churn your two favorite subjects? Then you are my BFF! But seriously, these two subjects combine in one great article...

AI companies have set unbelievable records for rapid growth in ARR, but these companies have equally astonishing churn rates.

#AIHype

sifted.eu/articles/ai-...
The AI bubble: fuelled by an ARR obsession, undone by churn?
ARR: the tech world’s favourite three-letter flex. LinkedIn is abuzz with discussion about which company has the highest.
sifted.eu
carl24k.bsky.social
I ❤️ #streamlit
carl24k.bsky.social
I say brace for my AI induced market volatility 😬
carl24k.bsky.social
This #AI report caused stocks to drop this week. Learnings & takeaways:

- Buying tools is better than in-house apps
- Back office automation most successful; most companies focus on sales and marketing.
- managers should drive adoption, not an AI group

#aihype

finance.yahoo.com/news/mit-rep...
MIT report: 95% of generative AI pilots at companies are failing
There’s a stark difference in success rates between companies that purchase AI tools from vendors and those that build them internally.
finance.yahoo.com
carl24k.bsky.social
And they might have to make some changes in the way productivity is measured, just as they adapted some GDP measurements to account for quality improvements.
carl24k.bsky.social
Nor should we be surprised that a lot of projects are cancelled. How many IT projects get cancelled? Lots, I assume. And it was probably even worse back in the 90's.
carl24k.bsky.social
If the "productivity paradox" of the 90's is any guide, we should start to see AI impacting productivity numbers in about 10 years. So not seeing AI "pay off" yet should not come as a surprise.

www.nytimes.com/2025/08/13/b...

#AI #LLM #AIHYPE
Companies Are Pouring Billions Into A.I. It Has Yet to Pay Off.
www.nytimes.com
carl24k.bsky.social
Yeah what’s a blue sky vibe anyway? But I do think it would be nice if there were more posts that expressed Manning’s character as an organization that supports learning, not just promoting specific products
carl24k.bsky.social
A Warm welcome to my publisher @manning.com who has joined us on Bluesky 🎉. #learntocode #learndatascience
carl24k.bsky.social
There are now academic theories of consciousness, most notably, the global workspace theory (GWS), and the integrative information theory (IIT) Maybe Claude can tell us which theory of consciousness would explain its own?
carl24k.bsky.social
This was a massive task, consuming a combined 3,000 hours of labor and a small army of 21 researchers! TheAgentCompany is meant to resemble a small software development company. The researchers made a number of tasks for the benchmarking — some easy, some hard, and some hard but with easy parts.