TGTGInsighttelegram intelligenceLIVE / telegram public index
← Midjourney/Nano Banana Prompts NeuroSpark

TGINSIGHT SIMILAR POSTS

Trouver du contenu similaire

Chaîne source @Shutter · Post #4607 · 22 mai

Harbor, cargo port, ships #AI#artificial_Intelligence

Résultats

10,064 posts similaires trouvés

Recherche globale générale

Learn RCRussian🤍💙❤️

@learnRCRussian · Post #6117 · 31/01/2026 16:20

Just in case you'd like to know what I'm going to do at the weekend. • Чипсы (plural noun) [chip-sy] = • Чипсики (dimin. plural) [chip-si-ki] Chips, crisps ❓️A у вас какие планы? 🐱Прекрасных выходных всем, товарищи! Обняла! #useful_vocabulary #just_cats #AI 🟠RCR | Support | Boost

Indian Development News 🇮🇳

@developmentnewsindia · Post #44332 · 02/05/2026 11:31

SAS has transformed its Indian operations into a primary global engineering hub with over 1,000 engineers driving innovation for its Viya platform and AI-driven risk, fraud, and compliance solutions. This move reflects the broader evolution of Indian Global Capability Centres (GCCs) into strategic "nerve centres" that command half of the global GCC workforce, with over 1,700 centers generating $65B+ in revenue and projecting growth toward $100B+ by 2030 #Ai#MakeinIndia

LIHAI 分享

@lihaiba · Post #6842 · 28/07/2024 03:41

▎🌐 秘塔AI搜索 一个暂无广告的搜索引擎 https://metaso.cn/ 合作/投稿频道群组 标签:#网站#AI

Hashtags

Parallel Experiments

@LinghaoCh · Post #938 · 19/04/2025 05:31

https://arxiv.org/abs/2305.18290#llm#ai 今天深入学习了 DPO,再次感叹扎实的数学功底对 AI/ML Research 的重要性…… 原始的 RLHF 是用 pairwise human preference data(A 和 B 哪个更好)去训练一个 reward model,然后用 RL 来训练主 policy model,objective 是 minimize negative log likelihood + regularization(比如 PPO 就是通过新旧 policy 之间的 KL Divergence 来做 regularization)。这样的缺点在于 RL 是出了名的难搞,而且还需要一个 critic model 来预测 reward,使得整个系统的复杂性很高。 DPO 的思路是,观察到 RLHF 的 objective 本质上是 minimize loss over (latent) reward function,通过一番 reparameterization 等数学推导,重新设计了一个 minimize loss over policy 的 objective,绕过了中间这个 reward model,让 gradient update 直接增加 policy model 生成 winner response 的概率并降低 loser response 的概率,大幅简化了流程。 拓展阅读: - KTO: 更进一步,不需要 pairwise comparison,只用对 individual example 的 upvote/downvote 也可以学习到 preference。 - IPO: 解决 DPO 容易 overfit 的问题。

Hashtags

12•••50•••9293949596•••100•••150•••200•••250•••300•••350•••400•••450•••500•••550•••600•••650•••700•••750•••800•••838839