SenseNova 5.0 Takes on GPT‑4 Turbo and Other AI Breakthroughs This Week
This roundup covers SenseTime's new SenseNova 5.0 model rivaling GPT‑4 Turbo, Apple's ReALM model that outperforms GPT‑4, the free‑to‑try Meshy 3 3D generator, Lamini's $25 M funding for enterprise generative AI, and OpenAI's upcoming ChatGPT‑powered search engine challenging Google.
SenseTime launches SenseNova 5.0
SenseTime announced the "日日新 SenseNova 5.0" large model with 600 billion parameters. In benchmark tests its inference, coding, and mathematics abilities match or exceed OpenAI's GPT‑4 Turbo and surpass Meta's Llama‑3 70B. The model uses a mixture‑of‑experts (MoE) architecture, a 200K context window, and over 10 TB of training tokens. Multi‑modal performance leads global rankings on MMBench and other benchmarks. Online demo: https://chat.sensetime.com/wb/home . Platform: https://platform.sensenova.cn/home . After release, SenseTime’s stock jumped over 30% before a trading halt and later settled up 17.5%.
Apple unveils ReALM model
Apple researchers introduced ReALM, a model that converts screen information into language‑model inputs, improving AI assistants' ability to handle ambiguous or complex queries. Tests show ReALM surpasses GPT‑4 on several metrics, with 1 B and 3 B parameter versions beating GPT‑4 on three benchmarks. Paper: https://arxiv.org/abs/2403.20329 .
Meshy 3 brings free Chinese 3D generation
Meshy 3, released on May 1, offers a free‑to‑try Chinese‑enabled text‑to‑3D service. The generated models support 360° high‑resolution textures, physical lighting, and can be exported as FBX, OBJ, GLB, or USDZ. Demo: https://app.meshy.ai/zh/discover .
Lamini raises $25 M for enterprise generative AI
Startup Lamini secured $25 million in a Series A round led by Amplify Partners. The company provides a platform for customizing large models with proprietary data and deploying generative AI applications. Its products include a memory‑evaluation suite, multi‑node training, JSON‑guaranteed output, and parameter‑efficient fine‑tuning. Customers include AMD, Snowflake, Databricks, and Meta. New funding will expand the team and GPU cloud infrastructure.
OpenAI plans a ChatGPT‑powered search engine
According to a leak, OpenAI has created the subdomain search.chatgpt.com and aims to launch a ChatGPT‑integrated search engine on May 9, directly challenging Google. The move coincides with Google I/O on May 14. Analysts note Google’s search suffers from ad overload, dead links, limited semantic understanding, personalization bubbles, and privacy concerns, while competitors like Microsoft’s Bing (now Copilot) and Perplexity AI have shown strong growth with LLM‑enhanced search.
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.
