Jolly

632 posts

Jolly

Jolly

@PowergentechX

เข้าร่วม Kasım 2025
327 กำลังติดตาม11 ผู้ติดตาม
Jolly
Jolly@PowergentechX·
I built a flight simulator where you can take part in the next war as it unfolds. A-10 Cuba Launches next week. #3js #cuba
Jolly tweet media
English
0
0
0
5
Jolly
Jolly@PowergentechX·
Hey Ai I want to go to Spain…. “That’s a great idea, first let’s drive to the beach. Next you’ll want to start wading into the ocean heading due east…. I’m drowning… Ah the classic “my lungs are filling with water”. Don’t worry, you’re almost there.
English
0
0
0
8
Jolly รีทวีตแล้ว
Elon Musk
Elon Musk@elonmusk·
Matter, Energy & Intelligence
English
12.7K
13.4K
110.9K
52.4M
Jolly รีทวีตแล้ว
Jolly รีทวีตแล้ว
Nick shirley
Nick shirley@nickshirleyy·
🚨 Here is the full 40 minutes of my crew and I exposing California fraud, Minnesota was big but California is even bigger... We uncovered over $170,000,000 in fraud as these fraudsters live in luxury with no consequences. Like it and share it, the fraud must STOP. We ALL work way too hard and pay too much in taxes for this to be happening. These fraudsters have been able to defraud American taxpayers for years without any pushback from the public and politicians. It is time to EXPOSE IT ALL and end America's fraud crisis.
English
12.2K
108.6K
333.5K
37.6M
Jolly รีทวีตแล้ว
will whang🌻
will whang🌻@will_whang·
A single photo that explain memory shortages /s
will whang🌻 tweet media
English
18
59
1.4K
70.3K
Jolly รีทวีตแล้ว
Aakash Gupta
Aakash Gupta@aakashgupta·
I worked at Epic Games for two years. This is real, and the strategy behind it is smarter than most people realize. Tim Sweeney has spent nearly two decades buying North Carolina forest land. 50,000+ acres across 15 counties. He’s now one of the largest private landowners in the state. The purchases started in 2008, right after the real estate collapse wiped out developers who had been planning golf resorts and luxury communities on biodiverse wilderness. Sweeney paid $15 million for Box Creek Wilderness, a 7,000-acre stretch in the Blue Ridge foothills containing 130+ rare and threatened species. Developers had owned 5,000 of those acres before the crash. He bought them for conservation prices when nobody else was bidding. He runs the acquisitions through an LLC called “130 of Chatham.” He buys the land, holds it for years, then either donates it to the U.S. Fish and Wildlife Service, sells it at a discount to state parks, or hands it to land trusts. In 2021, he donated 7,500 acres in the Roan Highlands to the Southern Appalachian Highlands Conservancy. Largest private land donation in North Carolina history. The part people miss: he told the News & Observer that since 2021, land got too expensive to keep buying. So he shifted focus to converting his existing 50,000 acres into permanent conservation status. He’s locking the land into legal structures that make development impossible regardless of who owns it in the future. A billionaire worth roughly $6 billion is spending tens of millions acquiring wilderness specifically during economic downturns, then giving it away or placing it under permanent legal protection. The land will outlast him, Epic Games, and Fortnite. That’s the part that separates Sweeney from billionaires who write checks to get their name on a building. The building depreciates. The forest compounds.
Dudes Posting Their W’s@DudespostingWs

Huge W

English
1.2K
19.2K
179.4K
10.5M
Jolly รีทวีตแล้ว
Perplexity
Perplexity@perplexity_ai·
Announcing Personal Computer. Personal Computer is an always on, local merge with Perplexity Computer that works for you 24/7. It's personal, secure, and works across your files, apps, and sessions through a continuously running Mac mini.
English
1.6K
3.5K
32.5K
14M
Jolly รีทวีตแล้ว
NYPD NEWS
NYPD NEWS@NYPDnews·
Some heroes wear capes — Chief Aaron Edwards wears blue.
NYPD NEWS tweet media
English
2.9K
11.5K
106K
26.6M
Jolly
Jolly@PowergentechX·
@karpathy @awnihannun youtu.be/5S3JgSAM_64?t=… Updating the brains mental is the most logical reason why sleep persists despite no obvious evolutionary benefit. I really like Jim Keller’s take on prepping for bedtime. Also explains why religions learned to pray before bed.
YouTube video
YouTube
English
0
0
0
0
Andrej Karpathy
Andrej Karpathy@karpathy·
There was a nice time where researchers talked about various ideas quite openly on twitter. (before they disappeared into the gold mines :)). My guess is that you can get quite far even in the current paradigm by introducing a number of memory ops as "tools" and throwing them into the mix in RL. E.g. current compaction and memory implementations are crappy, first, early examples that were somewhat bolted on, but both can be fairly easily generalized and made part of the optimization as just another tool during RL. That said neither of these is fully satisfying because clearly people are capable of some weight-based updates (my personal suspicion - mostly during sleep). So there should be even more room for more exotic approaches for long-term memory that do change the weights, but exactly - the details are not obvious. This is a lot more exciting, but also more into the realm of research outside of the established prod stack.
Awni Hannun@awnihannun

I've been thinking a bit about continual learning recently, especially as it relates to long-running agents (and running a few toy experiments with MLX). The status quo of prompt compaction coupled with recursive sub-agents is actually remarkably effective. Seems like we can go pretty far with this. (Prompt compaction = when the context window gets close to full, model generates a shorter summary, then start from scratch using the summary. Recursive sub-agents = decompose tasks into smaller tasks to deal with finite context windows) Recursive sub-agents will probably always be useful. But prompt compaction seems like a bit of an inefficient (though highly effective) hack. The are two other alternatives I know of 1. online fine-tuning and 2. memory based techniques. Online fine-tuning: train some LoRA adapters on data the model encounters during deployment. I'm less bullish on this in general. Aside from the engineering challenges of deploying custom models / adapters for each use case / user there are a some fundamental issues: - Online fine-tuning is inherently unstable. If you train on data in the target domain you can catastrophically destroy capabilities that you don't target. One way around this is to keep a mixed dataset with the new and the old. But this gets pretty complicated pretty quickly. - What does the data even look like for online fine tuning? Do you generate Q/A pairs based on the target domain to train the model? You also have the problem prioritizing information in the data mixture given finite capacity. Memory based techniques: basically a policy for keeping useful memory around and discarding what is not needed. This feels much more like how humans retain information: "use it or lose it". You only need a few things for this to work: - An eviction/retention policy. Something like "keep a memory if it has been accessed at least once in the last 10k tokens". - The policy needs to be efficiently computable - A place for the model to store and access long-term memory. Maybe a sparsely accessed KV cache would be sufficient. But for efficient access to a large memory a hierarchical data structure might be beter.

English
273
300
4.6K
574.9K
Jolly รีทวีตแล้ว
@amuse
@amuse@amuse·
ROBOTS? Graffiti artists assumed their jobs were safe. Then the guys working in the German Tesla factory decided they could just get robots to tag the factory walls.
English
36
193
1.5K
63.7K
Jolly รีทวีตแล้ว
Elon Musk
Elon Musk@elonmusk·
You can now select topics for your timeline, so if you’re tired of political ragebait, you can select something else 😂 Just tap on For You.
English
18.8K
19.3K
311.5K
84.1M
Jolly รีทวีตแล้ว
ib
ib@Indian_Bronson·
I asked Claude to look up the current situation and write a note to @DarioAmodei Dario, I say this as a friend: you are making a catastrophic strategic error, and the reasoning behind it doesn’t survive contact with reality. Your two red lines — no mass surveillance of Americans, no autonomous weapons without human-in-the-loop — sound principled in a vacuum. But you are not operating in a vacuum. You are operating in a world where the PLA is integrating AI into every layer of its kill chain with zero such scruples, where Chinese military AI development has no institutional review board, no congressional oversight, no ACLU, and no Dario Amodei demanding terms of service compliance. The practical effect of your stand is not that autonomous weapons don’t get built. They get built — by China, by Russia, by anyone not constrained by your moral framework. The practical effect is that the one military that actually has democratic accountability, civilian oversight, courts, a free press, and a functioning inspector general is the one that fights the next war with worse tools. You are not preventing dystopia. You are ensuring that if dystopia comes, it will be imposed by actors who never had to negotiate with you at all. Consider the logic chain: 1.You pull Claude from classified systems. 2.The Pentagon scrambles to Grok or Gemini — inferior models by everyone’s admission, including DoD’s own people. 3.The capability gap between the US and China widens in domains where AI is decisive: cyber, ISR fusion, targeting, logistics optimization. 4.The probability of a successful defense of Taiwan, or deterrence of a move on Taiwan, decreases. 5.The liberal democratic order you claim to value loses its security guarantor. You’ve told me yourself that you believe frontier AI is among the most consequential technologies in human history. If you actually believe that, how can you justify ensuring the US military — the only force standing between liberal democracy and its rivals — fields second-best AI? On what moral calculus does that work out? The Pentagon isn’t asking you to help build Skynet. They’re asking you to not have veto power over how a democratically accountable military uses a tool it purchased. Their point about “all lawful purposes” is actually the correct institutional boundary: the military operates under law, under civilian control, under congressional oversight. Your acceptable use policy is a private company substituting its judgment for the entire apparatus of democratic military governance. That’s the actual God complex here. The surveillance concern is a red herring in this context. The NSA already has authorities and tools for surveillance that dwarf anything Claude enables. You’re not preventing mass surveillance by withholding Claude — you’re just ensuring that whatever AI the government does use for those purposes is less safe, less auditable, and less aligned than yours. Same logic applies to autonomous weapons. Autonomous systems are coming regardless. The question is whether they’re built on a foundation that has your safety research baked in, or on something hacked together by a defense contractor with none of your alignment work. You are selecting for the worse outcome. I know you’re getting praised right now by exactly the people you’d expect. That praise is worth nothing when the strategic balance shifts and there’s no one left to protect the system that allows companies like Anthropic to exist in the first place. You are sacrificing the security of the civilization that makes your principles possible, in the name of those principles.
ib tweet media
English
174
645
3.8K
1.5M
Jolly รีทวีตแล้ว
Breaking911
Breaking911@Breaking911·
TRUMP: "If you agree with this statement then stand up and show your support: the first duty of the American government is to protect American citizens not illegal aliens."
English
303
1.2K
7.6K
655.9K