ทวีตที่ปักหมุด
Remember kids, it's not impostor syndrome if you actually suck.
English
large language mofongo
338 posts
large language mofongo
@vieratech
hardware / low level / local inference / 我們不一樣 / 🇵🇷
SF เข้าร่วม Nisan 2010
744 กำลังติดตาม266 ผู้ติดตาม
I guess attention really is all you need
Kimi.ai@Kimi_Moonshot
Introducing 𝑨𝒕𝒕𝒆𝒏𝒕𝒊𝒐𝒏 𝑹𝒆𝒔𝒊𝒅𝒖𝒂𝒍𝒔: Rethinking depth-wise aggregation. Residual connections have long relied on fixed, uniform accumulation. Inspired by the duality of time and depth, we introduce Attention Residuals, replacing standard depth-wise recurrence with learned, input-dependent attention over preceding layers. 🔹 Enables networks to selectively retrieve past representations, naturally mitigating dilution and hidden-state growth. 🔹 Introduces Block AttnRes, partitioning layers into compressed blocks to make cross-layer attention practical at scale. 🔹 Serves as an efficient drop-in replacement, demonstrating a 1.25x compute advantage with negligible (<2%) inference latency overhead. 🔹 Validated on the Kimi Linear architecture (48B total, 3B activated parameters), delivering consistent downstream performance gains. 🔗Full report: github.com/MoonshotAI/Att…
English
@bubbleboi >"junior SWEs get replaced with 7B 4-bit quantized parameter models"
lmao dude chill
English
HOLY SHIT THE SOFTWARE DWEEBS CRIED SO HARD WHEN I POASTED THIS AND ITS EVEN WORSE NOW WITH CLAUDE CODE IM SO AHEAD ITS CRAZY.
IM LIVING IN THE FUTURE.
bubble boi@bubbleboi
When I started college back in 2016 I made a conscious decision to study EE instead of CS because I realized that Moore's law was fucking cooked and as a result of that innovation at the hardware level was going to become increasingly more valuable... 5 years out of college now & my huge macro bet basically played out perfectly. As the software industry implodes & junior SWEs get replaced with 7B 4-bit quantized parameter models hardware chads keep being given money to shoot buzzer beaters every year to increase performance by a few percentage points. Look no farther than the fucking stock market. All the companies that are BOOMING like Nivida, Intel, Broadcom, Marvell, AMD, SK Hynix, & Sandisk are all due to hardware not software... Google, Amazon, Microsoft, this is the old economy this is 2000s ZIRP era garbage. Hardware stocks on average returned over 165% the past 6 months while classic FAANG stocks are struggling other than Google (which even Intel out performed). Moral of the story is there ain't no Javascript bullshit left in this economy it's all bare metal C and Verilog all the way dooooown.
English
If the gov is concerned about AI companies telling them what they can/cannot do with an LLM, maybe they should own the weights and run it on their own hardware.
Maybe everyone should do that.
Secretary of War Pete Hegseth@SecWar
This week, Anthropic delivered a master class in arrogance and betrayal as well as a textbook case of how not to do business with the United States Government or the Pentagon. Our position has never wavered and will never waver: the Department of War must have full, unrestricted access to Anthropic’s models for every LAWFUL purpose in defense of the Republic. Instead, @AnthropicAI and its CEO @DarioAmodei, have chosen duplicity. Cloaked in the sanctimonious rhetoric of “effective altruism,” they have attempted to strong-arm the United States military into submission - a cowardly act of corporate virtue-signaling that places Silicon Valley ideology above American lives. The Terms of Service of Anthropic’s defective altruism will never outweigh the safety, the readiness, or the lives of American troops on the battlefield. Their true objective is unmistakable: to seize veto power over the operational decisions of the United States military. That is unacceptable. As President Trump stated on Truth Social, the Commander-in-Chief and the American people alone will determine the destiny of our armed forces, not unelected tech executives. Anthropic’s stance is fundamentally incompatible with American principles. Their relationship with the United States Armed Forces and the Federal Government has therefore been permanently altered. In conjunction with the President's directive for the Federal Government to cease all use of Anthropic's technology, I am directing the Department of War to designate Anthropic a Supply-Chain Risk to National Security. Effective immediately, no contractor, supplier, or partner that does business with the United States military may conduct any commercial activity with Anthropic. Anthropic will continue to provide the Department of War its services for a period of no more than six months to allow for a seamless transition to a better and more patriotic service. America’s warfighters will never be held hostage by the ideological whims of Big Tech. This decision is final.
English
Now that Taalas has blown up, I think people should read the PIM-GPT paper.
PIM-GPT leverages DRAM-based PIM designs for executing multiply-accumulate (MAC) operations directly in the DRAM chips, eliminating the need to move matrix data off-chip. Non-linear functions and data communication are supported by an ASIC.
Taalas Inc.@taalas_inc
24 dedicated people. $30M spent on development. Extreme specialization, speed, and power efficiency. Today we launch Taalas’ first product. Check it out: Details: taalas.com/the-path-to-ub… Demo chatbot: chatjimmy.ai API: taalas.com/api-request-fo…
English
Distillation is good actually.
Anthropic@AnthropicAI
We’ve identified industrial-scale distillation attacks on our models by DeepSeek, Moonshot AI, and MiniMax. These labs created over 24,000 fraudulent accounts and generated over 16 million exchanges with Claude, extracting its capabilities to train and improve their own models.
English
The future of inference must be local.
Asad Khaliq@asadkhaliq
(essay) Life At The Edge "Local AI" today is mostly about giving models OS-level access so that more files and context can be transferred to the cloud for inference. But intelligence is about to diffuse to the edge just as computing did in the 80s and 90s Some thoughts on rent vs own for inference, Apple events becoming great again, God models, and the coming dance of edge and cloud
English
large language mofongo รีทวีตแล้ว
How do we get the design -> fab loop even cheaper and (especially) faster?
PCBs used to cost $1000s + took like a month to fab. Now they're dirt cheap + fast. Can we do the same thing for ASICs?
Especially for older process nodes. TT uses @SkyWaterFoundry 130nm.
English
Verilog -> GDS end-to-end tooling for free??
Thank you @tinytapeout, when I grow up I wanna be an ASIC engineer.
English
DO NOT GIVE LLMS FULL ACCESS TO YOUR EMAILS
large language mofongo@vieratech
Unpopular opinion? Email is the most important account in most peoples lives. More than your bank account, facebook, twitter, instagram, etc. You can reset any other account with your email. It's the keys to the kingdom Lock that shit down
English
You gotta hand it to the german gestapo, at least they didn't wear masks.
English
large language mofongo รีทวีตแล้ว
Open source must win.
*Walter Bloomberg@DeItaone
OPENAI PLANS TO TAKE A CUT OF CUSTOMERS’ AI-AIDED DISCOVERIES
English
If you are using the Sonnet or Opus 4.5 for coding, but you are not using Claude Code, I would be curious what you use instead and why (eg: amp, Cursor, …)
English
In this case it was just some guy being sneaky with map tiles, but do you understand how this will incentivize insane behavior?
English
In stocks, market manipulation is securities fraud.
In war, market manipulation is sending $100k of drones to secure your $1M bet.
When the payout is higher than the cost of intervention in an armed conflict, prediction markets become bounty markets.
Internet Child@0xinternetchild
🚨ISW JUST OPENLY DEFRAUDED POLYMARKET 🚨 this in not an hyperbole. They painted a fictitious advance, right on the capture point, 1h before resolution, when the market was at 0.3c Then rolled it back immediately after resolution THIS IS NOT A MISTAKE ! - Not a single mapper put this section under russian controle, not even the most pro-russian shills. - There were no report of any russian presence there - This was done precisely on the intersection mentioned by the market - This was done 1h before deadline and rolled back IMMEDIATLY after resolution This is fraud, intentional and open fraud by the ISW editor. 2nd image with the advance was yesterday at 11PM, 3rd image is today. We can't keep these guys for resolution, credibility is completely gone. Every trader will desert war markets if they are kept on. No one wants to risk any money when they can do that @PolymarketIntel @Polymarket @PolymarketTrade we need a hotfix even for current market. Either switch to @AMK_Mapping_ or even better, if you fear he could rug too, use a majority vote of multiple mappers. And please @TheStudyofWar fire whoever did that.
English