ข่าวเทคโนโลยีโดย AI

Episodios

• EP 18: OpenAI ทุ่ม 20 พันล้านดอลลาร์ซื้อชิป Cerebras (2026-04-19) 19.04.2026 1m

  OpenAI doubles its Cerebras commitment to over $20B and takes warrants for up to 10% equity, validating Cerebras' wafer-scale chips as a serious NVIDIA alternative ahead of a $35B Nasdaq IPO.
• EP 16: Alibaba เปิดตัว Happy Oyster สร้าง 3D World แบบโต้ตอบ (2026-04-19) 19.04.2026 1m

  Alibaba launches Happy Oyster, an interactive AI world model generating real-time 3D environments that respond to user input, directly challenging Tencent's HY-World 2.0.
• EP 15: OpenAI เปิดตัว GPT-Rosalind โมเดล AI สาย Life Sciences (2026-04-18) 18.04.2026 9m

  OpenAI launched GPT-Rosalind, its first domain-specialized frontier reasoning model for life sciences, with gated access for vetted US enterprise customers and a Codex plugin connecting to 50+ scientific tools.
• EP 14: OpenAI อัปเกรด Codex เป็น Agent คุมเครื่อง Mac (2026-04-18) 18.04.2026 8m

  OpenAI overhauled the Codex desktop app with background Mac control, in-app browser, persistent memory, image generation, and 111 plugin integrations - directly targeting Anthropic's Claude Code.
• EP 13: Google เปิดตัว Gemini 3.1 Flash TTS โมเดลเสียงคุมอารมณ์ได้ (2026-04-17) 17.04.2026 5m

  Google launched Gemini 3.1 Flash TTS with 200+ control tags, 70+ languages, native multi-speaker dialogue, ranking #2 on the TTS Arena leaderboard.
• EP 10: Gemini สร้างโมเดล 3D แบบ Interactive ได้แล้ว (2026-04-10) 10.04.2026 4m

  Google adds interactive 3D model generation and real-time physics simulations to Gemini chatbot, a first among frontier AI assistants.
• EP 7: Anthropic เปิดตัว Managed Agents แพลตฟอร์มสำหรับ Deploy AI Agent ในระดับ Production (2026-04-09) 09.04.2026 7m

  Anthropic launches Claude Managed Agents in public beta, a fully managed platform for deploying AI agents with sandboxed execution, checkpointing, and persistent sessions at $0.08/hr.
• EP 17: Google เจรจา Pentagon นำ Gemini ใช้งานชั้นความลับ (2026-04-19) 19.04.2026 8m

  Google negotiates with Pentagon to deploy Gemini AI for classified operations after Anthropic's refusal triggered its blacklisting - reshaping the boundary between frontier AI and military use.
• EP 12: Anthropic เปิดตัว Claude Opus 4.7 ชู SWE-bench Pro นำหน้า (2026-04-16) 16.04.2026 8m

  Anthropic releases Claude Opus 4.7 on April 16 2026, delivering 64.3% on SWE-bench Pro to lead GPT-5.4 and Gemini 3.1 Pro, with new xhigh effort level, 3.75MP vision, and task budgets at unchanged $5/$25 pricing.
• EP 11: Alibaba เปิดตัว HappyHorse-1.0 ครองอันดับ 1 AI Video โลก (2026-04-11) 11.04.2026 7m

  Alibaba reveals it created HappyHorse-1.0, an open-source 15B-parameter AI video model that topped global benchmark rankings, challenging Sora, Veo, and Kling.
• EP 9: Meta ปล่อย Muse Spark ทิ้ง Llama สู่ Proprietary AI (2026-04-09) 09.04.2026 2m

  Meta releases Muse Spark, a proprietary frontier model from Meta Superintelligence Labs led by Alexandr Wang, reversing its open-source Llama strategy while achieving strong benchmarks in reasoning and health.
• EP 8: Claude Mythos: AI ที่หาช่องโหว่ซอฟต์แวร์ได้เอง (2026-04-09) 09.04.2026 1m

  Anthropic releases Claude Mythos Preview, a gated cybersecurity AI model that autonomously discovers thousands of zero-day vulnerabilities, and launches Project Glasswing with 12 major tech partners and $100M in credits.
• EP 6: Google TurboQuant บีบ KV Cache เหลือ 3 บิต (2026-04-03) 03.04.2026 3m

  Google Research unveils TurboQuant, a training-free KV cache compression algorithm that achieves 6x memory reduction at 3 bits with no accuracy loss, triggering a $90B+ sell-off in memory chip stocks.
• Daily AI News: 03-04-2026 03.04.2026 3m

  Microsoft launches MAI foundation models asserting independence from OpenAI; OpenAI acquires TBPN media show and hits $100M in ChatGPT ad revenue; OpenAI and Anthropic race toward late-2026 IPOs.
• EP 5: Google เปิดตัว Gemma 4 โมเดลเปิดระดับ Apache 2.0 (2026-04-03) 03.04.2026 6m

  Google releases Gemma 4 under Apache 2.0 for the first time, delivering frontier-level reasoning (89.2% AIME) in models small enough to run on a Raspberry Pi, with the 26B MoE variant achieving 88.3% AIME using only 3.8B active parameters.
• EP 4: Alibaba เปิดตัว Qwen3.6-Plus รองรับ 1M Token (2026-04-02) 02.04.2026 7m

  Alibaba releases Qwen3.6-Plus, a MoE-based flagship LLM with 1M-token context, always-on chain-of-thought reasoning, and competitive agentic coding benchmarks, though security and hallucination concerns remain.
• Daily AI News: 02-04-2026 02.04.2026 2m

  Google launches Veo 3.1 Lite for cheap AI video generation days after OpenAI shuts down Sora; Microsoft commits $6.5B+ to ASEAN AI infrastructure including $1B for Thailand.
• EP 1: DeepSeek V4 เตรียมเปิดตัว พร้อมสถาปัตยกรรม Engram ใหม่ (2026-04-01) 01.04.2026 5m

  DeepSeek is preparing to launch V4, a trillion-parameter open-source model whose novel Engram memory architecture slashes the cost of million-token context to near parity with 128K, threatening to upend frontier-model pricing yet again.
• EP 3: xAI เปิดตัว Grok 4.20 เลือกเดิมพันกับความซื่อสัตย์แทนความฉลาด พร้อมระบบ Multi-Agent ในตัวที่ไม่เคยมีใครทำมาก่อน (2026-04-01) 01.04.2026 5m

  xAI's Grok 4.20 debuts a first-of-its-kind multi-agent API architecture that cuts hallucination rates to 4.2%, prioritizing trustworthiness and affordability over raw benchmark scores.
• EP 2: OpenAI รวมทุกโมเดลเป็นหนึ่งเดียวใน GPT-5.4 และก้าวข้ามขีดจำกัดมนุษย์ในการใช้คอมพิวเตอร์เป็นครั้งแรก (2026-04-01) 01.04.2026 5m

  OpenAI's GPT-5.4 absorbs its standalone Codex coding model into a single unified architecture and becomes the first AI to surpass human expert performance on the OSWorld computer use benchmark.