ycchen1808

[ycchen1808 @ycchen1808]

@nangongyuan 我第一次看到ai短劇有想回應些什麼的衝動，但又不知道該說什麼

[ycchen1808 @ycchen1808]

@Kawa_GameDev （5）效能和能力不符預期，看起來蒸餾被阻礙真的對開源模型影響很大

[ycchen1808 @ycchen1808]

@SamVale29 @grok 我是牙醫，何時會被取代？ @grok

[ycchen1808 @ycchen1808]

@NVIDIAGeForce #RTXON actually i need my third and fourth sparks

[NVIDIA GeForce @NVIDIAGeForce]

We're giving away a Hogwarts Legacy custom wrapped NVIDIA GeForce RTX 5090 to celebrate its official release with DLSS 5 🪄

To enter:
🟢 Share this post
🟢 Comment #RTXON

Terms and Conditions: https://t.co/45T3R732hf

[ycchen1808 @ycchen1808]

@MiaAI_lab 1017 THE BEST

[ycchen1808 @ycchen1808]

@HouseHackerJon opus

[ycchen1808 @ycchen1808]

@elonmusk @DogecoinNorway 為什麼grokbot那麼消耗資源？我覺得一個個人agent消耗的資源應該遠遠小得多，是因為運行的模型嗎？幾個b300就能提供絕大部分的人運行dav4.1-flash1017

[ycchen1808 @ycchen1808]

@cz_binance @grok 如果讓grok bot幫我開啟連結跟檢查附件能規避風險嗎？

[ycchen1808 @ycchen1808]

@elonmusk @cz_binance 讓grok bot幫我開呢？

[ycchen1808 @ycchen1808]

@julsix2333 真的吃完九個菜，我會怕妳沒吃飽

[ycchen1808 @ycchen1808]

@NFT_Chen 直接用無審查版不好嗎？而且這種prompt提示jailbreak deepseek很容易就改掉了

[Petrus Pennanen @petruspennanen]

AI and memory - so many solutions, so little progress.

How do you make a local LLM remember new facts, corrections, and deletions while serving - without touching its base weights or bloating every prompt?

So we got to work at TensorFold. @ashxhart made a revolutionary learning neuronal memory layer, and I added an exact learning layer of ngrams with promotion of the most important ones to neurons. And now we are presenting

TensorFold Memory: Fast to write, slow to learn - just like human memory.

With two layers that run live while serving:

TensorFold Notes: exact token-ID n-gram store tested up to 1M facts with ms lookup, instant edits and exact delete plus live lookup during decode.
Living Weights: promotes stable facts into ~10 extra neurons/fact in the background for 102.1 tok/s decode on GB10 - same as untaught

In our tests a few retrieved notes match putting every fact in the prompt and two-step reason much better than neurons alone. And it's not only for TensorFold - you can find a Notes Proxy in https://t.co/XlDa53XXsQ that adds Notes to inference engines such as llama.cpp, vLLM and Ollama.

Read the paper here: https://t.co/vDHFtALg8T

[轟@YouTube @shoto_afi]

ソフトバンクの孫社長が言っていた。
人生を変える手順は3つだけ。

1.脳がちぎれるほど考え抜く。
2.登る山を決める。
3.決めたら腹をくくる。

たったこれだけ。逆に最も無意味なのが、

[0x鸣人 @LuBtc888]

马斯克曾经也是个高内耗人格，

直到后来他找到了个屡试不爽的方法：

[ycchen1808 @ycchen1808]

@designwkarthick IPO,可能把原本用於模型訓練的算力撥給了推理吧

[ycchen1808 @ycchen1808]

@MiaAI_lab dgx spark完全被打敗了嗎?
還有價值嗎?

[ycchen1808 @ycchen1808]

@MiaAI_lab 人們會在睡前這麼對Mia說….

[ycchen1808 @ycchen1808]

@MiaAI_lab 量化損失多嗎?有沒有機會出現無審查版?

[ycchen1808 @ycchen1808]

@MiaAI_lab goodjob