Unique Research · Weekly AI Worldview · April 15, 2024
One week of global AI funding, events, and people. Charts attach on ingest (20 images).
This week’s headlines
Funding — MyTwins.ai multi-¥10M angel; NieTa (捏Ta) multi-¥10M; ModelBest (面壁智能) multi-$100M-RMB round with Primavera and Huawei Hubble.
Events — Microsoft £2.5B UK / London AI center; Tencent Cloud AIGC storage upgrade; Meta Llama 3 (~140B); Gemini 1.5 Pro upgrade; Microsoft $2.9B Japan; Meta 5nm AI chip; CodeFuse image-to-code; GPT-4 Turbo Vision API; Infosys × Intel; Google Axion; ARK into OpenAI; Adobe video training data; Sony BRC-AM7; Huawei MateBook X Pro + Pangu.
People — Andrew Ng joins Amazon board; Robin Li on closed-source Wenxin; Jordan Hoffman to lead Microsoft consumer AI London center.
Funding
- MyTwins.ai — multi-¥10M angel led jointly by Will Hunting Ventures (将门创投), Yongxi Asset (永禧资产), and Westlake Sci-Tech Venture (西湖科创投); Yuanqi Capital (元启资本) exclusive FA. General AI for personal digital twins — promotion, communication, and service upgrades. Self-developed multimodal AI enables real dialogue and motion; personalized twins plus industry foundation-model builds. Partnered with known streamers to lift efficiency and live duration; targeting education, healthcare, finance, marketing. Digital twins as consumer “virtual partners.” (MyTwins.ai)
- NieTa (捏Ta) — angel + Pre-A totaling over ¥10M, backed by Source Code Capital (源码资本) and MiraclePlus (奇绩创坛). Funds for model optimization, multimodal features, and social play. AIGC character-creation platform; team from senior internet backgrounds (incl. AIDXL). New app to upgrade community for multi-person, multi-character, multimodal interactive story creation and sharing. (NieTa)
- ModelBest (面壁智能) — new multi-$100M-RMB (数亿) round led by Primavera Capital (春华创投) and Huawei Hubble (华为哈勃), with Beijing AI Industry Investment Fund and others; Zhihu (知乎) also added. Funds for talent, foundation-model compute/data base, and application rollout. Core team from Tsinghua NLP Lab; previously shipped Chinese LLM CPM-1. Focused on AGI; landed in finance, education, government. Primavera: ModelBest shows forward vision and a full tech stack; strong data and talent base for foundation-model progress. (ModelBest)
Events
- Microsoft — £2.5B UK / London AI center — New London AI center led by Jordan Hoffman to advance consumer AI products and research and compete for talent, tapping the UK AI ecosystem. Over the next three years Microsoft plans to invest £2.5B in the UK to expand data centers and train AI workforce skills. May intensify talent competition with Google DeepMind and peers; Microsoft is confident the center can attract and grow talent. (Microsoft)
- Tencent Cloud AIGC storage upgrade — Full upgrade of AIGC-scene cloud storage: large-model data cleaning and training efficiency roughly doubled / time halved. Stack: object storage COS, high-performance parallel file storage CFS Turbo, data accelerator GooseFS, and Cloud Infinite (CI). COS supports hundred-EB scale with efficient public-net ingest. CFS Turbo after optimization: aggregate R/W throughput at TiB/s class; metadata performance at million OPS. MetaInsight product for multimodal data search. Storage evolving toward intelligent storage and data applications. (Tencent Cloud)
- Meta Llama 3 (~140B) — Meta preparing next-gen foundation model Llama 3 at roughly 140B parameters — high performance with cost efficiency for SMEs and individual developers. Series history: LLaMA open-sourced Feb last year; Llama 2 optimized for commercial use after ~2T tokens training and ~1M human-annotation fine-tunes — among the most widely used open GenAI models. Llama 3 expected to further push tech and applications. (Meta)
- Google Gemini 1.5 Pro upgrade — Auto movie-review writing and multimodal understanding; 1M-token context; available in 180+ countries/regions. Google AI Studio free trial with Chinese prompt support. Gemini API stability and output quality improved. (Google)
- Microsoft — $2.9B Japan — Up to $2.9B into Japan to expand data centers and introduce advanced semiconductor tech; strengthen domestic management/protection of personal and confidential data. With OpenAI, plans to offer ChatGPT foundation tech in Japan and set up a dedicated research base for robotics. Japan market ranks after U.S. and China; GenAI market projected to reach ¥30T JPY by 2030. (Microsoft) Note: CN title summarized as “$2.5B Japan”; body states $2.9B — body figure used.
- Meta 5nm AI chip — Developing AI chips via TSMC 5nm to cut external-supplier dependence; compute and memory bandwidth more than double. Shifting toward AI services with surging compute need; plans $35B AI infrastructure investment; previously bought hundreds of thousands of NVIDIA H100s for recommendation and GenAI. (Meta)
- Ant Group CodeFuse — image-to-code — Intelligent R&D platform CodeFuse ships “image-to-code” so front-end engineers generate code from product design mockups in one click (internal beta). Already in daily R&D: half of engineers adopt weekly; 10% of new code AI-generated; adoption rate 30%, rising to 50% in unit-test scenes. Built on Ant’s Ling (百灵) foundation model; covers full R&D chain. (Ant Group)
- OpenAI GPT-4 Turbo Vision API — Vision in GPT-4 Turbo API: image classification, detection, data conversion, and related ops — turning images, labels, symbols into structured data. Training data updated through December 2023; context up to 128K tokens. (OpenAI)
- Infosys × Intel — April 10, 2024 deepen strategic partnership around Infosys Topaz platform integrating Intel-based solutions to help clients embed GenAI into operations. (Infosys)
- Google Axion AI CPU — Arm-based high-performance CPU for data-center AI needs; aims to reduce dependence on NVIDIA and Amazon and raise AI-market competitiveness. Claims better performance than current Arm and x86 instances; also improving TPU chips. Plan to integrate Axion into Google Cloud. (Google)
- ARK Invest into OpenAI — Cathie Wood’s ARK disclosed its $54M venture fund holds OpenAI shares; OpenAI seen at the AI frontier. Via ARK closed-end fund framework that also holds SpaceX, Epic Games, etc. OpenAI already heavily backed by Microsoft; valuation cited at $86B. (ARK)
- Adobe video training data — Pushing video-generation R&D; offering photographers/artists $120 compensation for high-quality video used to train AI models; more detail later this year. Also buys stills when library is short, with contributor rates from $0.06–$0.16. (Bloomberg)
- Sony BRC-AM7 — April 11 AI auto-framing flagship PTZ camera: subject recognition and smooth auto-tracking; 20× optical zoom; 4K 60P; works with Sony imaging systems for meetings, broadcast, live events, sports. Planned launch early 2025. (Sony)
- Huawei MateBook X Pro + Pangu — April 11 new MateBook X Pro with Pangu foundation model; thin magnesium-alloy body; Intel Core Ultra 9 + Arc GPU; 40W performance release. AI summary, AI minutes, voice input, AI subtitles; first NPU-driven AI effects with Wondershare Filmora (万兴喵影); WPS AI tools (smart layout, conversational tables, mind-map→PPT); Super Device multi-device synergy. (Huawei)
People
- Andrew Ng (吴恩达) joins Amazon board — April 11 Amazon announced Ng joining the board; Judy McGrath will not continue as director — a signal of business pivot. Despite heavy video/music investment, McGrath not standing for the next board may imply adjustments; Amazon has already cut hundreds of roles and scaled back Prime Video. To stay tech-leading, Amazon is increasing AI investment; Ng’s experience is expected to help. (Amazon)
- Robin Li (李彦宏) on closed-source Wenxin — In a recent internal talk, Li first addressed why Wenxin stays closed: open models attract developers, but closed models will keep leading technically; closed mode concentrates compute and talent for better cost control and performance; at equal params closed models perform better. For AI startups he argued “dual-wheel drive” is not ideal — it can scatter resources and hurt core competitiveness. (Tencent News Frontline)
- Microsoft consumer AI + Jordan Hoffman — Microsoft forming a new consumer AI unit with a London center led by Jordan Hoffman (ex-Inflection AI; Inflection founded by DeepMind co-founder Mustafa Suleyman). New hiring to follow for language models plus related infra and tools. (Microsoft)
— END —