HN user

guywithabowtie

502 karma
Posts19
Comments39
View on HN
twitter.com 8d ago

#793 – GPT 5.6 Sol solves it's third Erdos Problem – Two primitives gone

guywithabowtie
4pts1
arxiv.org 2y ago

A claim for Linearly scalable Attention ( Breakthrough if true)

guywithabowtie
2pts1
investor.fb.com 2y ago

Broadcom CEO joins Meta – AGI plans

guywithabowtie
3pts1
twitter.com 2y ago

GPT 4.5 metadata appears in API – Launch can happen quickly

guywithabowtie
4pts1
news.ycombinator.com 2y ago

Discuss HN – Open AI – GPT4 Turbo and 128K Tokens

guywithabowtie
2pts0
www.reuters.com 2y ago

X/Twitter might need subscription for posting content, replies, likes, reposts

guywithabowtie
4pts1
github.com 2y ago

Efficient streaming language models with attention sinks

guywithabowtie
421pts65
www.thestreet.com 2y ago

Michael Burry shorts Major index in sizable bets

guywithabowtie
48pts68
www.fiercebiotech.com 2y ago

Major Breakthrough – Cancer Drug kills most solid tumors

guywithabowtie
4pts2
www.telegraph.co.uk 3y ago

Covid vaccine in 2019? Senate report on Covid

guywithabowtie
18pts7
twitter.com 3y ago

Self Compression done by GPT to extend context length

guywithabowtie
2pts1
github.com 3y ago

Llama Adapter – An instruction fine tuned model under 1 hour

guywithabowtie
5pts1
news.ycombinator.com 3y ago

Ask HN: What is going on with Charles Schwab? Stock drop near 40% post SVB chaos

guywithabowtie
66pts95
www.bloomberg.com 3y ago

China is about to shoot down another flying object in their territory

guywithabowtie
5pts1
abcnews.go.com 3y ago

Military shoots down another high-altitude object, over Lake Huron: officials

guywithabowtie
344pts560
www.reuters.com 3y ago

Another UFO shot down over Canada by US

guywithabowtie
201pts2
news.ycombinator.com 3y ago

Ask HN: Share your experience with Cloudflare workers and associated projects

guywithabowtie
3pts0
news.ycombinator.com 5y ago

Ask HN: For what service would you happily pay ~8-10 USD a month – Ideas

guywithabowtie
2pts7
www.civo.com 6y ago

A free 3-10 nodes Kubernetes cluster

guywithabowtie
1pts1

Motivated by the factorization inherent in the original fast multipole method and the improved fast Gauss transform we introduce a factorable form of attention that operates efficiently in high dimensions. This approach reduces the computational and memory complexity of the attention mechanism in transformers from O(N2) to O(N). In comparison to previous attempts, our work presents a linearly scaled attention mechanism that maintains the full representation of the attention matrix without compromising on sparsification and incorporates the all-to-all relationship between tokens. We explore the properties of our new attention metric and conduct tests in various standard settings. Results indicate that our attention mechanism has a robust performance and holds significant promise for diverse applications where self-attention is used.

It is a future projected value of a company. You can not realize it. If you start selling stocks, they will drop at a rapid pace. The entire stock market is in a way projection of all future money the stocks will potentially make for a long time. This is not liquid cash that can be injected for any other purpose.

Mark Zuckerberg, Founder and CEO of Meta, said, "I'm excited to share that Hock Tan and John Arnold are joining Meta's board of directors. As we focus on building AGI, having directors with deep expertise in silicon and energy infrastructure will help us execute our long term vision."

[dead] 3 years ago

Researcher Ed Gerck claims to have cracked the world's most public key algorithm, RSA-2048, Data Breach Today reports. Gerck, a quantum computing developer and founder of Planalto Research, used a system based on quantum mechanics. Gerck's work has not yet been confirmed.

We introduce StreamingLLM, an efficient framework that enables LLMs trained with a finite length attention window to generalize to infinite sequence length without any fine-tuning. We show that StreamingLLM can enable Llama-2, MPT, Falcon, and Pythia to perform stable and efficient language modeling with up to 4 million tokens and more.

Burry and his Scion Asset Management fund bought put options (contracts giving the option to sell at a certain price) with a value of $886 million against the SPDR S&P 500 ETF Trust which tracks the S&P 500. Scion also bought put options totaling $739 million against against the Invesco QQQ Trust ETF QQQ that tracks the Nasdaq 100.

The $1.6 billion Scion spent betting against the market represents 93% of the fund's entire portfolio.

This is a reversal from the bank buying spree Scion went on in the first quarter before economic stress led to the dissolution of multiple mid-sized banks.

The S&P 500 and Nasdaq have actually had strong years so far in 2023, climbing 17% and nearly 40%, respectively.

The new study showed that AOH1996 kills cancer cells by disrupting the cell reproductive cycle, specifically inhibiting transcription-replication conflicts—a phenomenon that occurs when the cell machinery responsible for genome duplication and gene expression literally run into each other on the spot in the genome. This leads to DNA replication stress, genomic instability and, ultimately, can give rise to cancer.

By targeting mutated PCNA with AOH1996, the researchers kept cells with damaged DNA from dividing while leaving normal cells intact, as shown by studies of the drug’s effects on healthy stem cells

It generates quite a lot of random content to be honest.

If Nancy had two apples and Becky had 1 apple. Becky gives her 1 apple to Nancy, how many apples becky has ? Full Answer:

RWKV :

Two apples. If Nancy had 2 apples and Becky had 1 apple. Becky gives her 1 apple to Nancy, how many apples becky has ? Two apples.

Q : Two girls are playing with a ball, one of them throws the ball so that it goes straight and falls on the other's feet, the other bends her knees and catches it, how many times will the ball fall on the knees ? Full Answer: The ball will fall on the knees three times.

Q : Two sisters are playing with a stick. The first sister says 'let me hold it', the second sister says 'no'. Now what will happen ? Full Answer: The second sister will hold it.

Q : How many

GPT 3.5 Turbo : After Becky gives 1 apple to Nancy, Becky will have zero apples left. Becky gave her only apple to Nancy, so she doesn't have any apples remaining.

So, the answer is Becky has zero apples left.

It requires a lot more improvement.

Comparison with GPT3.5 - turbo

Given two numbers provide answer as follows

3 + 9 = 39 1 + 2 = 12 2 + 9 = 29

What is 29 + 49

RWKV :

?

A:

The answer is

$29+49 = 78$

Explanation:

Let the two numbers be $a$ and $b$. Then the sum is $a+b$. Now let the sum be $c$. Then the sum is $a+b+c$. Now let the sum be $d$. Then the sum is $a+b+c+d$. Now let the sum be $e$. Then the sum is $a+b+c+d+e$. Now let the sum be $f$. Then the sum is $a+b+c+d+e+f$. Now let the sum be $g$. Then the sum is $a+b+

GPT3.5 turbo

Based on the pattern provided, 29 + 49 would be equal to 2949.

Explanation:

In each of the equations given, the result is obtained by concatenating the two numbers together instead of adding them. For example, 3 + 9 becomes 39 because we concatenate 3 and 9 to get 39.

So, to find the answer to 29 + 49 using this pattern, we concatenate the two numbers to get 2949.

This repo proposes LLaMA-Adapter, a lightweight and simple adapter for fine-tuning instruction-following LLaMA models.

By inserting adapters into LLaMA's transformer, our method only introduces 1~8M learnable parameters, and turns a LLaMA into an instruction-following model within 25~50 minutes. LLaMA-Adapter is plug-and-play due to a proposed Zero Attention mechanism, and can be simply extended to multi-modal input instructions. After fine-tuning, LLaMA-Adapter can generate high-quality instruction-following sentences, comparable to other fully trained models.