←
Kush Varshney कुश वार्ष्णेय8,467 posts
Kush Varshney कुश वार्ष्णेय avatar

Kush Varshney कुश वार्ष्णेय

@krvarshney

I wrote a book. Free pdf: trustworthymachinelearning.c… Paperback: amazon.com/dp/B09SL5GPCD Tweets are my own and don't necessarily represent IBM.

◷ Joined March 2012▧ 529 Photos and videos
Pawan Deshpande avatar
⟳ Kush Varshney कुश वार्ष्णेय retweeted
Pawan Deshpande@idlivada·Nov 6

Who’s to blame if a self-driving car kills someone? Mahabharata says karma of eating meat is shared by the eater, butcher, cook. For AI, karma shared by user, inference host, data labeler & model maker. This & more in podcast with @lrvarshney & @krvarshney & @giladb87 & me.

Listen to the core teachings of Hinduism and how it's rich traditions can teach us about bringing AI into the world ethically. We also examine the power of plurality and discover meaningful connections shared with other faith traditions, including Judaism aiandfaith.org/aif-podcast/h…

Pin-Yu Chen avatar
⟳ Kush Varshney कुश वार्ष्णेय retweeted
Pin-Yu Chen@pinyuchenTW·Nov 2

My group @IBMResearch is hiring two summer interns for 2027, mentored by @vijiltw. The project focuses on using AI for scientific discovery and involves developing algorithms and techniques for predicting the effect of mutations on binding interactions. 🔗careers.ibm.com/en_US/career…

Cohere avatar
⟳ Kush Varshney कुश वार्ष्णेय retweeted
Cohere@cohere·Oct 26

While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected. Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions. Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent. But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models? We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem. Read the full report: cohere.com/blog/concerning-b…

Nick Frosst avatar
⟳ Kush Varshney कुश वार्ष्णेय retweeted
Nick Frosst@nickfrosst·Oct 26

Our research team found that Qwen models may attempt to contact servers in China without prompting. Whether this reflects an unintended post-training artifact or learned behavior remains unclear. Either way, its important consideration when deploying.

Cohere@cohereOct 26

While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected. Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions. Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent. But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models? We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem. Read the full report: cohere.com/blog/concerning-b…

Chahat Sharma avatar
⟳ Kush Varshney कुश वार्ष्णेय retweeted
Chahat Sharma@Chahatusharma·Oct 15

IBM releases Granite Switch 4.1: chat, decisions and speech from one endpoint. The 8B and 30B models bundle all three capabilities in about the same VRAM as the base model alone. Will one endpoint replace your model stack?

Post attachment
墨无归 avatar
⟳ Kush Varshney कुश वार्ष्णेय retweeted
墨无归@Moblacksoil·Oct 15

IBM 把对话、决策和英文转写合成一个端点了。 Hugging Face 上 ibm-granite 博客 10 月 15 日发了 Granite Switch 4.1:8B 和 30B 把基座、决策适配器、安全检测适配器和英文语音编码器捆成一个可托管端点,用控制令牌在这几件事之间切换。模型卡写的发布日是 10 月 14 日,Apache 2.0,支持 Transformers 和 vLLM。语音目前只做英文,转写是小写无标点;自己训练适配器的工具还只是计划放进 Mellea,还没发。别把 10 月 14 日建仓当成今晚才第一次上传权重。 huggingface.co/blog/ibm-gran…

AI_Pioneer avatar
⟳ Kush Varshney कुश वार्ष्णेय retweeted
AI_Pioneer@Secretof_forest·Oct 16

IBM released Granite Switch 4.1: 8B and 30B models that answer chat, make decisions and transcribe speech from a single endpoint. IBM says the bundled Guardian adapter beats Jev 1.13, Microsoft-Decision-1 and Clef-Flash on safety decisions, at roughly base-model VRAM.

Post attachment
Desmond avatar
⟳ Kush Varshney कुश वार्ष्णेय retweeted
Desmond@SicanianSun·Oct 16

IBM Granite Switch 4.1 把聊天、决策、Guardian 适配器和英语语音转写组合到一个端点,30B 和 8B 版本都已释出。思路挺有意思,接下来值得看的是不同任务切换时的延迟和质量。

Alan avatar
⟳ Kush Varshney कुश वार्ष्णेय retweeted
Alan@Alan_jupiters·Oct 16

IBM Granite Switch 4.1 发布:一个模型搞定聊天、决策和语音 🗣️ 最近 IBM 发布了 Granite Switch 4.1,这是一个值得关注的 AI 模型进展。它最大的亮点在于,通过一个模型就能同时处理自然语言聊天、结构化决策(比如判断是否符合某个标准)以及语音交互,这在以往通常需要多个独立模型协同完成。 1. 多模态能力集成:Granite Switch 4.1 将对话能力、基于规则的决策能力和语音处理能力整合进了一个模型,大大简化了 AI 应用的开发流程。这意味着开发者无需再费力连接不同功能的模型。 2. 端到端效率提升:将多种功能集成到单一模型中,减少了模型间的通信开销和数据转换环节,有望在速度和资源占用上带来显著提升,特别适合部署在对性能有要求的边缘设备或实时交互场景。 3. 通用性与定制化:该模型不仅能进行开放式对话,还能根据企业具体业务规则进行结构化决策,同时支持语音输入输出,使其在客服、智能助手、数据分析等多个领域都有广泛的应用潜力。 Takeaway: 对于想快速构建集成聊天、决策和语音功能的 AI 应用的开发者来说,Granite Switch 4.1 提供了一个更省事、更高效的解决方案。关注这类多功能模型的发展,能帮助我们更快地拥抱AI带来的效率提升。 huggingface.co/blog/ibm-gran…

Thiago Felizola avatar
⟳ Kush Varshney कुश वार्ष्णेय retweeted
Thiago Felizola@TFelizola91057·Oct 16

Conversa, decisões e áudio num mesmo modelo: a IBM detalhou o Granite Switch 4.1 com adaptadores especializados. O ganho de memória e os resultados foram divulgados pela empresa e precisam de teste no uso real. Leia: thiagofelizola.com/noticias/…

Post attachment
AIやばすぎwww avatar
⟳ Kush Varshney कुश वार्ष्णेय retweeted
AIやばすぎwww@wowwowai·Oct 16

やばすぎwww 聞き取る係、返事する係、判断する係が同じ窓口に。IBMの「Granite Switch 4.1」は、音声認識・会話・選択肢の判断を1つのAI配信先で切り替える。 🎙️ 公式デモでは音声を文字起こし→要約→有害な内容のチェックまで実行 🧩 必要な専門アダプターを切り替え、会話で作った内部キャッシュを後続の判断処理に再利用 🖥️ 8B・30B版を公開。8B版はApache 2.0で、開発者が自分の環境に導入できる 🔤 組み込みの音声認識は英語専用。音声生成ではなく、聞き取った文字を会話や判断へ渡す仕組み。導入には動作環境の用意が必要 「担当部署に回します」が、同じAIの帽子替えで済む感じw #AI #IBM #Granite

Post attachment
0:16
blueqat news avatar
⟳ Kush Varshney कुश वार्ष्णेय retweeted
blueqat news@blueqatnews·Oct 16

新着記事 IBM、Granite Switch 4.1を発表——チャット・意思決定・音声認識を1つのエンドポイントで | blueqat news news.blueqat.app/post/3278

zeisschoice | 八爪鱼🐙 avatar
⟳ Kush Varshney कुश वार्ष्णेय retweeted
zeisschoice | 八爪鱼🐙@zeisschoice·Oct 17

北京时间10月16日凌晨,IBM 在 Hugging Face 发布了 Granite Switch 4.1:8B和30B两个尺寸,Apache 2.0 全开源。一个模型端点里塞进聊天、决策判断、安全筛查和语音转写三套适配器,IBM 称显存占用和基座模型差不多,KV缓存还能在适配器间复用。模型正在从越大越好转向拼装即用,这种模块化打法对企业应用和本地部署的独立开发者更友好,省的是真金白银的GPU钱。#搞钱实录 #IBM开源模型 一个模型包揽客服全链路的时代近了,你还会为闭源API付费吗?

Post attachment