Carlos Rangel

395 posts

Carlos Rangel banner
Carlos Rangel

Carlos Rangel

@cloudmkv

参加日 Temmuz 2025
43 フォロー中15 フォロワー
Carlos Rangel
Carlos Rangel@cloudmkv·
@TukiFromKL Thats because we're telling it "btw im a crypto guy" at the start of every conversation.
English
0
0
0
48
Tuki
Tuki@TukiFromKL·
🚨 Are you paying attention to what Karpathy just admitted.. the founding member of OpenAI.. the guy who trained the models you use every day.. just said every single LLM has the same problem.. ask it one question two months ago and it treats it like your entire identity.. oh wait.. you mentioned crypto once in January? congratulations.. you're now a crypto guy forever.. you also asked about a recipe? every conversation starts with "as someone who enjoys cooking.." these models don't remember you.. they stereotype you.. off a single data point.. we gave AI a photographic memory and forgot to give it the ability to forget.. and forgetting is the most human thing there is..
Andrej Karpathy@karpathy

One common issue with personalization in all LLMs is how distracting memory seems to be for the models. A single question from 2 months ago about some topic can keep coming up as some kind of a deep interest of mine with undue mentions in perpetuity. Some kind of trying too hard.

English
246
328
4.4K
1.1M
Carlos Rangel
Carlos Rangel@cloudmkv·
@karpathy Yep I agree, it's technically in the same realm as few shotting, and its so bad. It's best to minimize the initial context load (claude.md/agents/memory) to only what you really need, or perhaps (implementation detail) have the human review / optimize at start of new threads
English
0
0
0
10
Andrej Karpathy
Andrej Karpathy@karpathy·
One common issue with personalization in all LLMs is how distracting memory seems to be for the models. A single question from 2 months ago about some topic can keep coming up as some kind of a deep interest of mine with undue mentions in perpetuity. Some kind of trying too hard.
English
1.5K
907
18.1K
2.1M
Carlos Rangel
Carlos Rangel@cloudmkv·
@eliano If you make a deep dark blue teal one I would buy 10
English
0
0
0
65
Carlos Rangel
Carlos Rangel@cloudmkv·
@ADHD_Capital He is 100000% correct. Entry level workers of most positions are the first to go. If you’re in charge of someone, chances are you are now responsible for doing their work but with AI now. Less cogs, more decision makers.
English
2
0
1
322
Carlos Rangel
Carlos Rangel@cloudmkv·
@peterwildeford No, this isn’t continual learning, it’s more akin to having notes you can look back to but you always have to look back to them. Updating weights is real continual learning. This is just context.
English
0
0
3
117
Carlos Rangel
Carlos Rangel@cloudmkv·
@vitrupo Agents can do anything anyone on a computer, especially a terminal, can do. Not shocking
English
0
0
0
32
vitrupo
vitrupo@vitrupo·
Andrej Karpathy told his OpenClaw: “I think I have Sonos at home.” It scanned the network, found devices, reverse engineered APIs, and started playing music. Soon “Dobby” was running the house. Lights. HVAC. Security cameras. Agents are getting good at figuring systems out.
English
107
155
2.4K
346.8K
Carlos Rangel
Carlos Rangel@cloudmkv·
@MeekMill meek you got adhd bro youre one of us this isn't an insult btw
English
0
0
1
216
Carlos Rangel
Carlos Rangel@cloudmkv·
@oguzerkan I would be shocked if this isn't already priced in. This is why Jensen was pushing the idea of token factories. It boils down to how many tokens can you produce per watt of electricity you use.
English
0
0
0
351
Oguz Erkan
Oguz Erkan@oguzerkan·
You aren’t bullish enough on $AMZN. This got 30 million views in just 11 hours. It gives everybody a complete remote worker in their pocket. You don’t need to deal with MCP, connectors, or anything else. Can you imagine what this will do to token usage? Models can interact with other applications also powered by LLMs using your computer. Token usage will skyrocket, so will Anthropic’s revenues. AWS will be a direct beneficiary as most of Anthropic’s inference workloads run on AWS. Barclays predict AWS growth to hit 34% in Q3 this year while UBS sees a path for 38% annual growth. You aren’t bullish enough.
Claude@claudeai

You can now enable Claude to use your computer to complete tasks. It opens your apps, navigates your browser, fills in spreadsheets—anything you'd do sitting at your desk. Research preview in Claude Cowork and Claude Code, macOS only.

English
41
31
826
226K
StockStorm
StockStorm@StockStormX·
@aleabitoreddit Admitting $AXTI is overvalued at $4B and then saying someone would pay $10B for it
English
10
0
5
3.6K
Serenity
Serenity@aleabitoreddit·
$AXTI has now reached the legendary status of $69.69. Was expecting close to 10x returns in a year and half, not a few months… Everything from $AAOI to $SIVE are ridiculously outperforming. As for AXT valuations at $4B? Is it overvalued. Yes. Would hyperscalers pay $10B to secure their AI buildout? Yes. Lot of these bottlenecks can’t be valued with traditional metrics.
Serenity tweet mediaSerenity tweet media
English
38
33
522
281.7K
Lars Christensen
Lars Christensen@MaMoMVPY·
A year ago Jensen was a completely reasonable sounding CEO. Now he and the rest of the AI CEOs are just making up obvious fantasies about the state of AI. We have not achieved AGI. He knows that we are nowhere near that.
Yuchen Jin@Yuchenj_UW

Jensen: “I think we’ve achieved AGI.” “It’s not out of the question that an OpenClaw could create a web service or some interesting little app that, all of a sudden, a few billion people use for 50 cents.”

English
137
214
2.6K
94.3K
Todd Saunders
Todd Saunders@toddsaunders·
Claude is going to be the biggest software company ever. Every day I feel like my mind is blown. They can now control your mouse, keyboard, and screen. Combine that with Claude Code, MCP connectors and dispatch for remote execution…. They are essentially building a universal employee that operates every single piece of software your company uses. But my biggest question heading into the AI future is, what happens to your product/service when humans stop clicking? The company that controls the workflow/clicks controls procurement, and that’s Anthropic.
Felix Rieseberg@felixrieseberg

Today, we’re releasing a feature that allows Claude to control your computer: Mouse, keyboard, and screen, giving it the ability to use any app. I believe this is especially useful if used with Dispatch, which allows you to remotely control Claude on your computer while you’re away.

English
32
17
253
68.1K
Carlos Rangel
Carlos Rangel@cloudmkv·
@oguzerkan token factories have been widely pushed by jensen, its definitely being priced in already, just anticipating growth signals and continued growth in capex to sustain current ai stock values tbh
English
0
0
0
229
Oguz Erkan
Oguz Erkan@oguzerkan·
Buy every cloud provider and every chip maker, we probably need at least 100x the compute we currently have. Scaling computer use will skyrocket the token consumption. AI models can use other AI-powered apps through your computer, creating very long sessions of token consumption across different applications and models. These sessions can easily consume multiples of tokens a human controlled session could consume. The market hasn’t priced scaling of this type of usage. Very bullish for cloud providers and chip makers. $NBIS $AMZN $MSFT $NVDA $AMD
Claude@claudeai

You can now enable Claude to use your computer to complete tasks. It opens your apps, navigates your browser, fills in spreadsheets—anything you'd do sitting at your desk. Research preview in Claude Cowork and Claude Code, macOS only.

English
17
10
209
41.4K