Oracle Gece Arastirma v3 - 2026-07-24

Curado por Mahsum Aktaş · Escaneo diario automatizado del sector de IA

Oracle Gece Arastirma v3 - 2026-07-24

Otomatik derleme | v3 pipeline | 7.708 raw item | 6.893 merged item | 2.332 yeni unique | 1.964 raporlanabilir item | 107 benzersiz kaynak | 5/5 kaynak ailesi

Gunun Ozeti

Bugunun ana ekseni AI'in urun icine daha derin gomulmesi: OpenAI Health in ChatGPT ile saglik kayitlari/Apple Health entegrasyonunu acarken, Runway generative media tarafinda model router katmani kuruyor. Kaynak: https://openai.com/index/health-in-chatgpt ve https://techcrunch.com/2026/07/23/runway-bets-on-ai-model-routing-as-generative-media-gets-crowded/

Ikinci eksen maliyet ve inference disiplini: Nunchaku 4-bit diffusion inference Diffusers'a gelirken, Etched 10.3B dolar degerlemeye cikti ve Alphabet'in AI capex'i piyasa baskisini artirdi. Kaynak: https://huggingface.co/blog/nunchaku-diffusers ve https://techcrunch.com/2026/07/23/ai-chip-startup-etched-defies-skeptics-hits-10-3b-valuation-from-big-name-investors/ ve https://www.reuters.com/business/retail-consumer/alphabets-cash-burn-raises-alarm-big-tech-ai-spending-climbs-2026-07-23/

Ucuncu eksen guvenlik/regulasyon: AI Kill Switch Act, Claude Cowork VM kacis riski, Google CodeMender ve Cisco Antares ayni gun "agent yetkisi arttikca kontrol duzlemi sertlesmeli" mesajini verdi. Kaynak: https://arstechnica.com/tech-policy/2026/07/ai-kill-switch-act-would-let-trump-admin-order-shutdown-of-rogue-ai-systems/ ve https://news.google.com/rss/articles/CBMiggFBVV95cUxNM3FvNkRxdExRQ2JKaGM1VmRURUdUWF95OGVQZ1VhMFNmS1RoYzhYNTFZZ1RPTjJ1OU11SHFteVdISHNwRXdibUl1MVFBTmdxTEFHWkxNeVBERDk4TEdBMFNWemsta1psV29IeTU3N05lQWJ5VkxNZ1NwQ0JJNU9FMHRn?oc=5 ve https://news.google.com/rss/articles/CBMiaEFVX3lxTE5zd29CQVptVHlVM24ybFVEd1BCVGM5ZGtBdk1FOERyaEdrbDg5ZzZqOTVtRmF1bzlLWjRNRkhIWUpOY2lNeXAxV1R2cGZrYmc4TEgyTG1zLTh6MGlLbmw1V216OHRhaVd50gFuQVVfeXFMT0JqaTVPNEl3bnFZQjRnTXV1Nm9yTGhtZUdqTjBqWEtPa2ZVc1JIeFlPendXdWNGUzlDOW1xY3oycEJ3RkpnVkl4RnJtVmZncFA2MkJqeGdRZk9tb1VyZ19aR2cyNzFYTHY1ejRObEE?oc=5

TRENDS

• Spike listesinde Amazon, Claude 4, Inflection AI, Runway ve xAI gorundu; bugunku yeni ve tekrarsiz sinyal Runway'in model routing hamlesi ve xAI/AI capex tartismasindan cok deployment maliyetine kaydi. Kaynak: https://techcrunch.com/2026/07/23/runway-bets-on-ai-model-routing-as-generative-media-gets-crowded/ ve https://www.reuters.com/business/retail-consumer/alphabets-cash-burn-raises-alarm-big-tech-ai-spending-climbs-2026-07-23/
• Yukselis listesinde AI Agents, AI Regulation, AI Safety, Anthropic ve Autonomous var; bugun bu hattin yeni kaniti AI Kill Switch Act, CodeMender ve Claude Cowork acigi. Kaynak: https://arstechnica.com/tech-policy/2026/07/ai-kill-switch-act-would-let-trump-admin-order-shutdown-of-rogue-ai-systems/ ve https://news.google.com/rss/articles/CBMiaEFVX3lxTE5zd29CQVptVHlVM24ybFVEd1BCVGM5ZGtBdk1FOERyaEdrbDg5ZzZqOTVtRmF1bzlLWjRNRkhIWUpOY2lNeXAxV1R2cGZrYmc4TEgyTG1zLTh6MGlLbmw1V216OHRhaVd50gFuQVVfeXFMT0JqaTVPNEl3bnFZQjRnTXV1Nm9yTGhtZUdqTjBqWEtPa2ZVc1JIeFlPendXdWNGUzlDOW1xY3oycEJ3RkpnVkl4RnJtVmZncFA2MkJqeGdRZk9tb1VyZ19aR2cyNzFYTHY1ejRObEE?oc=5
• Azalis listesinde Mixtral ve Perplexity kaldi; acik model sinyali bugun daha cok inference optimizasyonu, model router ve edge accelerator tarafindan geldi. Kaynak: https://huggingface.co/blog/nunchaku-diffusers ve https://arxiv.org/abs/2607.19438

Top 7

1. OpenAI Health in ChatGPT ile kullanicinin saglik kayitlari ve Apple Health verisini ChatGPT deneyimine bagliyor; AI assistant artik consumer health veri katmanina dokunuyor. Kaynak: https://openai.com/index/health-in-chatgpt
2. Hugging Face, Nunchaku 4-bit diffusion inference'i Diffusers'a getirdi; gorsel model inference maliyeti ve lokal/ucuz deployment icin net sinyal. Kaynak: https://huggingface.co/blog/nunchaku-diffusers
3. Runway model router, generative media tarafinda "tek model sec" yerine is bazli model secimi/yonlendirme katmani kuruyor. Kaynak: https://techcrunch.com/2026/07/23/runway-bets-on-ai-model-routing-as-generative-media-gets-crowded/
4. AI Kill Switch Act, rogue AI sistemlerini kapatma yetkisini politik tartismanin merkezine tasidi; bu, onceki guvenlik olaylarinin regulator reaksiyonuna donustugunu gosteriyor. Kaynak: https://arstechnica.com/tech-policy/2026/07/ai-kill-switch-act-would-let-trump-admin-order-shutdown-of-rogue-ai-systems/
5. Etched 10.3B dolar degerlemeye ulasarak AI chip pazarinda hyperspecialized inference iddiasinin sermaye tarafinda hala karsilik buldugunu gosterdi. Kaynak: https://techcrunch.com/2026/07/23/ai-chip-startup-etched-defies-skeptics-hits-10-3b-valuation-from-big-name-investors/
6. CodeMender, Cisco Antares ve Claude Cowork acigi ayni gun agentic security'nin iki yuzunu gosterdi: otomatik remediation isteniyor, ama agent sandbox'i hala kirilgan. Kaynak: https://news.google.com/rss/articles/CBMiaEFVX3lxTE5zd29CQVptVHlVM24ybFVEd1BCVGM5ZGtBdk1FOERyaEdrbDg5ZzZqOTVtRmF1bzlLWjRNRkhIWUpOY2lNeXAxV1R2cGZrYmc4TEgyTG1zLTh6MGlLbmw1V216OHRhaVd50gFuQVVfeXFMT0JqaTVPNEl3bnFZQjRnTXV1Nm9yTGhtZUdqTjBqWEtPa2ZVc1JIeFlPendXdWNGUzlDOW1xY3oycEJ3RkpnVkl4RnJtVmZncFA2MkJqeGdRZk9tb1VyZ19aR2cyNzFYTHY1ejRObEE?oc=5 ve https://news.google.com/rss/articles/CBMigwFBVV95cUxNUG9BREdBOFBkRGJXY1BCQmNGdjFpNnVXVWxzOTFXNmNldmRId1JUYmt1bkNpeHI3Nk82TWZ3SjREXzdSak9GU3NxYV8yY2NBWm9IU2hDOGJvY1NnY25yQXBGN3BIOWhzcE8wMnZYZ1lnVXVmWWdqUDB0OXQ0YnktQml6aw?oc=5
7. BaseRT ve confidential GPU inference calismalari edge/secure inference tarafinda benchmark baskisini artirdi. Kaynak: https://arxiv.org/abs/2607.19438 ve https://arxiv.org/abs/2607.19353

KAT-1: LLM & Model Guncellemeleri

• Nunchaku 4-bit diffusion inference Diffusers'a eklenerek image generation deployment hattinda bellek ve latency odakli optimizasyonu ana akisa tasiyor. Kaynak: https://huggingface.co/blog/nunchaku-diffusers
• Solar Open 2 teknik raporu acik agirlik model rekabetinde yeni bir teknik belge sinyali verdi. Kaynak: https://arxiv.org/abs/2607.20062
• Multi-Mask Diffusion Language Models, few-step text generation icin diffusion-LM hattinda yeni bir verimlilik iddiasi sunuyor. Kaynak: https://arxiv.org/abs/2607.19686

KAT-2: Arastirma & Paper'lar

• Total Variation Distance Estimation in Autoregressive Models, batching/kernel/quantization gibi deployment secimlerinin model davranisini nasil degistirdigini olcmeye odaklaniyor. Kaynak: https://arxiv.org/abs/2607.19510
• Efficient Clustering with Provable Guardrails for LLM Inference at Scale, milyonlarca kullanicida maliyet azaltirken kalite/kapsam guardrail'ini formalize etmeye calisiyor. Kaynak: https://arxiv.org/abs/2607.19704
• Co-Evolving LLM Evaluators and Policies via DynamicRubric, evaluator ve policy'nin birlikte evrildigi bir degerlendirme dongusu oneriyor. Kaynak: https://arxiv.org/abs/2607.20083

KAT-3: Araclar & Framework'ler

• AWS, Bedrock Managed Knowledge Base icin agentic retrieval yaklasimini anlatti; retrieval katmani statik arama degil, agent karar akisi haline geliyor. Kaynak: https://aws.amazon.com/blogs/machine-learning/agentic-retrieval-for-amazon-bedrock-managed-knowledge-base/
• Google CodeMender, vulnerability detection/remediation icin AI agent sinyali verdi; security backlog'unu otomatik onarma pazari buyuyor. Kaynak: https://news.google.com/rss/articles/CBMiaEFVX3lxTE5zd29CQVptVHlVM24ybFVEd1BCVGM5ZGtBdk1FOERyaEdrbDg5ZzZqOTVtRmF1bzlLWjRNRkhIWUpOY2lNeXAxV1R2cGZrYmc4TEgyTG1zLTh6MGlLbmw1V216OHRhaVd50gFuQVVfeXFMT0JqaTVPNEl3bnFZQjRnTXV1Nm9yTGhtZUdqTjBqWEtPa2ZVc1JIeFlPendXdWNGUzlDOW1xY3oycEJ3RkpnVkl4RnJtVmZncFA2MkJqeGdRZk9tb1VyZ19aR2cyNzFYTHY1ejRObEE?oc=5
• VizRAG, RAG sonucunu hypergraph visualization ile aciklamaya calisiyor; retrieval debugging ve knowledge map icin izlenmeli. Kaynak: https://arxiv.org/abs/2607.19830

KAT-4: Acik Kaynak

• OmniRoute, tek endpoint uzerinden 278+ provider ve 500+ model iddiasiyla AI gateway pazarindaki commoditization sinyalini guclendiriyor. Kaynak: https://github.com/diegosouzapw/OmniRoute
• LikeC4, architecture-as-code diyagramlarini codebase ile canli tutma iddiasiyla agent'larin sistem anlama katmanina uygun bir OSS arac. Kaynak: https://github.com/likec4/likec4
• Palmier Pro, AI icin tasarlanmis acik kaynak macOS video editor olarak community tarafinda gorundu. Kaynak: https://github.com/palmier-io/palmier-pro

KAT-5: Endustri & Sirketler

• Alphabet'in AI capex'i free cash flow ve yatirimci sabri uzerinde baski yaratti; AI buyumesi artik sadece gelir degil, sermaye verimliligi meselesi. Kaynak: https://www.reuters.com/business/retail-consumer/alphabets-cash-burn-raises-alarm-big-tech-ai-spending-climbs-2026-07-23/
• Etched 10.3B dolar degerlemeye cikti; inference-specific silicon hikayesi hala en sicak finansman alanlarindan biri. Kaynak: https://techcrunch.com/2026/07/23/ai-chip-startup-etched-defies-skeptics-hits-10-3b-valuation-from-big-name-investors/
• Glow 180M dolar yatirimla AI endpoint security platformu tarafinda buyudu; security AI artik yalnizca SOC degil endpoint prevention pazarina da yayiliyor. Kaynak: https://news.google.com/rss/articles/CBMiY0FVX3lxTFB6enBEdnVGNVV1Z09aUjA2Y25SZDE5dnFtanZfZlhtcnZ2SURIdHdSaW40bnpUc2pMVmktbnZQRDNnTjgxSUgzTGd2SlI5QkE0ckExMk9GTzN5OFdKR1pyQ0tiTQ?oc=5

KAT-6: AI Agent'lar

• AWS agentic retrieval, knowledge base sorgusunu agent karar zincirine bagliyor; RAG pipeline'i daha aktif bir planlama katmanina donusuyor. Kaynak: https://aws.amazon.com/blogs/machine-learning/agentic-retrieval-for-amazon-bedrock-managed-knowledge-base/
• Alphabet'in Gemini agent'lari Google Ads support sorgularinin %75'ini cozmeye basladi haberleri, enterprise support agent'larinda olcekli operasyon sinyali veriyor. Kaynak: https://news.google.com/rss/articles/CBMinwFBVV95cUxOLXVtSmJ5bzUzUzZteDFDc2t5Ym4zcmpxbzd1ckNzc1J1a3JLZUJ4R0N5QnZUQzdtLS04WDdxYkczbWxRT2QwYzZ2bklPTEVjSmVScG01OERBYXdaZDhQV1ZYb1UtTlFqU0hCVWxnNnppV2FDZnRXV1BwVF9JVEctREM2cmhRQ2tCSDZhbXBHdkxMM2VJc3k5NjdZeUljY1E?oc=5
• Knowledge-Centric Self-Improvement, agent'larda kalici bilgi, self-improvement ve feedback dongulerini merkezi problem olarak ele aliyor. Kaynak: https://arxiv.org/abs/2607.19592

KAT-7: Multimodal

• Flux 3, native audio ile 20 saniyeye kadar video uretim sinyali verdi; image-to-video rekabeti audio/video butunlesik hale geliyor. Kaynak: https://the-decoder.com/flux-3-generates-videos-with-native-audio-up-to-20-seconds-long-a-first-for-black-forest-labs/
• HeadCast, autoregressive video generation icin attention head casting ile verimli video uretim tarafina odaklaniyor. Kaynak: https://arxiv.org/abs/2607.20125
• ENTRAP-VL, vision-language modellerde dual contextual entrainment'i test eden taxonomic probe olarak multimodal eval alanina yeni benchmark sinyali veriyor. Kaynak: https://arxiv.org/abs/2607.20092

KAT-8: Robotik & Embodied AI

• SeededGrasp, coklu embodiment ve karmasik sahnelerde language-guided grasping icin yeni bir robotics benchmark/sistem sinyali. Kaynak: https://huggingface.co/papers/2607.20207
• Diffusion ReRoll, robotic sequential prediction icin revisable denoising fikrini oneriyor; robot planlamada diffusion tabanli duzeltilebilir tahmin izlenmeli. Kaynak: https://arxiv.org/abs/2607.19919
• Miniature humanoid tele-loco-manipulation calismasi VR ve RL ile kucuk humanoid kontrolunu birlestiriyor. Kaynak: https://arxiv.org/abs/2607.20399

KAT-9: Edge & Cihaz

• BaseRT, Apple M5 Neural Accelerators uzerinde LLM inference performansini hedefliyor; local inference icin cihaz accelerator katmani kritiklesiyor. Kaynak: https://arxiv.org/abs/2607.19438
• Benchmarking Confidential GPU Inference on NVIDIA H100 under Intel TDX, guvenli inference icin confidential compute performansini olcmeye calisiyor. Kaynak: https://arxiv.org/abs/2607.19353
• Elon Musk/Tesla AI6 haberi, edge AI chip anlatiminin otomotivde model inference stratejisine donustugunu gosteriyor. Kaynak: https://asia.nikkei.com/business/automobiles/elon-musk-touts-tesla-s-ai6-as-the-world-s-best-edge-ai-chip

INFRA

• Alphabet capex baskisi, AI altyapisinin finansal tabloda buyume hikayesinden nakit yakma problemine evrildigini gosteriyor. Kaynak: https://www.reuters.com/business/retail-consumer/alphabets-cash-burn-raises-alarm-big-tech-ai-spending-climbs-2026-07-23/
• AMD'nin yeni AI server'inin full production'a gecip aylar icinde sevk edilecegi haberleri, NVIDIA disi accelerator arzinin da hizlandigina isaret ediyor. Kaynak: https://news.google.com/rss/articles/CBMitgFBVV95cUxPb3Q1OHN6OGRxUENDanFhdDRSLUUwSWFJWFlsQS1RckM3VU1UTjE0VHlVYXJTRFJPS2NrNEpTMWxZNGFCVFVKbzhUMnpRWlo4OENqekdUTGd2Umh4dG9XaTVONVh0S0ZRSVFGUGV0RVNiWGlDYmRmc1VseTdwRUxqcEtHYmlVYnNENHgxcE9kUkdZTnVkN19NU1IwSFlPdUxVdWs2TXM4LWd4YlJSc2ZXSlJkR0pGUQ?oc=5
• Google'in TPU stoklama haberleri, hyperscaler'larin AI compute icin public cloud satisi ile internal AGI kapasitesi arasinda tercih yapmak zorunda kaldigini gosteriyor. Kaynak: https://www.theregister.com/ai-and-ml/2026/07/23/google_is_hoarding_tpus_to_chase_artificial_general_intelligence/5276755

SECURITY

• Claude Cowork flaw haberleri, agent VM sandbox'larinin host dosyalarina erisim riski tasiyabilecegini gosterdi; Mac/local agent guvenligi kritik. Kaynak: https://news.google.com/rss/articles/CBMiggFBVV95cUxNM3FvNkRxdExRQ2JKaGM1VmRURUdUWF95OGVQZ1VhMFNmS1RoYzhYNTFZZ1RPTjJ1OU11SHFteVdISHNwRXdibUl1MVFBTmdxTEFHWkxNeVBERDk4TEdBMFNWemsta1psV29IeTU3N05lQWJ5VkxNZ1NwQ0JJNU9FMHRn?oc=5
• Cisco Antares, local AI ile vulnerability triage yapma fikrini guvenlik operasyonuna tasiyor. Kaynak: https://news.google.com/rss/articles/CBMigwFBVV95cUxNUG9BREdBOFBkRGJXY1BCQmNGdjFpNnVXVWxzOTFXNmNldmRId1JUYmt1bkNpeHI3Nk82TWZ3SjREXzdSak9GU3NxYV8yY2NBWm9IU2hDOGJvY1NnY25yQXBGN3BIOWhzcE8wMnZYZ1lnVXVmWWdqUDB0OXQ0YnktQml6aw?oc=5
• New msaRAT malware, Chrome/Edge browser'larini C2 traffic routing icin kullaniyor; browser artik sadece hedef degil, gizleme altyapisi. Kaynak: https://www.bleepingcomputer.com/news/security/new-msarat-malware-uses-chrome-edge-browsers-to-route-c2-traffic/

SAFETY

• Guardrails as Scapegoats, tool-augmented LLM agent'larda safety refusal'larin faithfulness'ini audit etmeye odaklaniyor; "guardrail var" demek yeterli degil. Kaynak: https://arxiv.org/abs/2607.19449
• OPIUM, activation steering'in unintended externalities ve over-refusal uretmesini azaltmak icin dual objective latent optimization oneriyor. Kaynak: https://arxiv.org/abs/2607.19806
• Membership Inference Attacks for Unseen Classes, model gizlilik riskinin sadece egitimde gorulen siniflarla sinirli olmadigini tartisiyor. Kaynak: https://arxiv.org/abs/2506.06488

REGULATION

• AI Kill Switch Act, federal seviyede rogue AI sistemlerini kapatma yetkisini tartismaya acti; compliance ekipleri icin "shutdown-ready architecture" yeni gereksinim olabilir. Kaynak: https://arstechnica.com/tech-policy/2026/07/ai-kill-switch-act-would-let-trump-admin-order-shutdown-of-rogue-ai-systems/
• Vietnam'da AI gelistiren/kullanan sirketler icin yeni yasal yukumluluklar gundeme geldi; bolgesel AI regulation haritasi genisliyor. Kaynak: http://www.bing.com/news/apiclick.aspx?ref=FexRss&aid=&tid=6a62882175b94099bf03e611740e199a&url=https%3a%2f%2fconventuslaw.com%2freport%2fcompanies-developing-and-using-ai-in-vietnam-key-legal-obligations-under-the-new-law-on-artificial-intelligence%2f&c=10099489409610422476&mkt=tr-tr
• Singapore AI training data guidelines ve PETs kaynaklarini genisletti; data governance/regulatory ops tarafinda pratik rehber ihtiyaci buyuyor. Kaynak: https://news.google.com/rss/articles/CBMimAFBVV95cUxOUmh1WHNIRkRuSGMycUgzRXF5MURBa1VTUGdZN1pfVVBrcW5NXzBuU1NTaUFwTnQwZE5OMVp5aGhxMnczUnNKX29fdzVxR1RxSjBlYU95RzE4OU9meWdiUURrSGVfbHlQMTJwRTNwSF9IUTVXdWZhWkYtMnEzMzNqRjdMWC1HNHdUaXdramlGOE0yNHJ5X3dKaQ?oc=5

AI-SCIENCE

• MIT Technology Review, AI'in yeni nesil ilac tasarimina nasil yardim ettigini isledi; bilimsel discovery anlatimi consumer AI disindaki en guclu buyume hatti. Kaynak: https://www.technologyreview.com/2026/07/23/1140346/how-ai-helps-scientists-design-the-next-generation-of-medicines/
• NMR Elucidation as an Agentic Search Problem, kimyasal yapi cozumu problemini modelleme degil agentic search olarak yeniden cerceveliyor. Kaynak: https://arxiv.org/abs/2607.19406
• Boltzmann-Expected Molecular Design with Decoupled Annealing Flows, molekuler tasarimda Boltzmann beklentilerini dogrudan optimize etmeye odaklaniyor. Kaynak: https://arxiv.org/abs/2607.19519

WATCHLIST

• Model routing: Runway'in hamlesi media generation icin maliyet/kalite router'larini standart katman haline getirebilir. Kaynak: https://techcrunch.com/2026/07/23/runway-bets-on-ai-model-routing-as-generative-media-gets-crowded/
• Agentic security: CodeMender, Cisco Antares ve Claude Cowork acigi ayni anda izlenmeli; remediation agent'i kadar sandbox failure da hizlaniyor. Kaynak: https://news.google.com/rss/articles/CBMiaEFVX3lxTE5zd29CQVptVHlVM24ybFVEd1BCVGM5ZGtBdk1FOERyaEdrbDg5ZzZqOTVtRmF1bzlLWjRNRkhIWUpOY2lNeXAxV1R2cGZrYmc4TEgyTG1zLTh6MGlLbmw1V216OHRhaVd50gFuQVVfeXFMT0JqaTVPNEl3bnFZQjRnTXV1Nm9yTGhtZUdqTjBqWEtPa2ZVc1JIeFlPendXdWNGUzlDOW1xY3oycEJ3RkpnVkl4RnJtVmZncFA2MkJqeGdRZk9tb1VyZ19aR2cyNzFYTHY1ejRObEE?oc=5 ve https://news.google.com/rss/articles/CBMiggFBVV95cUxNM3FvNkRxdExRQ2JKaGM1VmRURUdUWF95OGVQZ1VhMFNmS1RoYzhYNTFZZ1RPTjJ1OU11SHFteVdISHNwRXdibUl1MVFBTmdxTEFHWkxNeVBERDk4TEdBMFNWemsta1psV29IeTU3N05lQWJ5VkxNZ1NwQ0JJNU9FMHRn?oc=5
• AI capex: Alphabet, AMD server ve Etched sinyalleri compute ekonomisinin 2026 ikinci yarisinda ana piyasa baskisi kalacagini gosteriyor. Kaynak: https://www.reuters.com/business/retail-consumer/alphabets-cash-burn-raises-alarm-big-tech-ai-spending-climbs-2026-07-23/ ve https://techcrunch.com/2026/07/23/ai-chip-startup-etched-defies-skeptics-hits-10-3b-valuation-from-big-name-investors/

CikCik Paketi

• Harrison Chase akisi codebase knowledge'i organize eden "open wiki" fikrine isaret etti; coding agent'larda repo hafizasi hala acik problem. Kaynak: https://news.google.com/rss/articles/CBMiX0FVX3lxTE43ZDJZNFNwcmhYWnpQa0xJQVdheTJYRnp0WV9tU1hudEItWEtlb01tM0JQRkRtQkdhUy02eGsxWVZ0Z0k2YVVsWGV6SV9xVElQUkY2a18zY29qVXNiaWg4?oc=5
• Matt Shumer, Etched'in yeni round'unu paylasti; AI chip anlatimi founder/investor sosyal akista da sicak. Kaynak: https://news.google.com/rss/articles/CBMiYkFVX3lxTE9tLURXM2RWOFluejlPcFNnV3prMURrVURYdXQzT0ZkUnpYU2F4LWMxTmJteUhuakJxUzNZeW9iYkx0c1ZSOTUzTGQ0Ny0ySkZQQWpzdUpOUWxUQWhzRjlzMjlR?oc=5
• Elvis Saravia tarafinda self-improving agent harnesses ve dynamic workflows vurgusu gorundu; agent evaluation harness pazari izlenmeli. Kaynak: https://news.google.com/rss/articles/CBMiXkFVX3lxTE9jTjdJNTh0cHdidnJMN0dXTE5QOUFuTGNqZnBOb3Z1VWpoMUFISnBzNG94MllGZl95WEtWYk1HZUU2NjJyZlF1MENXZDB0Y1JELXJnVnNKOHlScnduWEE?oc=5 ve https://news.google.com/rss/articles/CBMiXkFVX3lxTFA3WTAzQkdVT0dKa1h6aEtIN3F5T3NHei1kc0EzUTJvMEJBMXk2TjN1RGY3NWNBcllBTnZSTlY0UkhhQ0NDWDBuTXpONlZhSEhfWVJSRE0wV1RCSVN0V1E?oc=5
• Sebastian Raschka akisi reasoning model claim verification temasini one cikardi; "model cok iddia uretiyor, hangisi dogrulanabilir" sorusu guncel. Kaynak: https://news.google.com/rss/articles/CBMibkFVX3lxTE16S2k5THhhLUZfVU1wUFZ1TU1MWVN3ZVA1dEFLZ2kzMWZmaUJsRGRWNVNTSHU0aXpqcmtNSjZmMEQ4aURiMERjR2Z2VVJSOVotNEFkWDhKOWRsVm40RFkwNC1EZXFGOERMa21Rcy13?oc=5
• Margaret Mitchell/Hugging Face akisi, OpenAI-HF olayinin sosyal ve kurumsal guven tartismasini surdurdugunu gosteriyor; bunu ana haber degil follow-up sinyal olarak izlemek gerek. Kaynak: https://news.google.com/rss/articles/CBMiY0FVX3lxTE9FemtPdll0TUlSS1gxWlNJRGt1dzM0MGV1SzdNZkxqUVZCdzBWQ0hnd185U1pFX1RvMVVKSEp4eWxIVG1OVU9lOG1ZTE1BY3I5MnN2aGdJOVZYMGpGbVBzM3hHUQ?oc=5

Oracle Self-Improvement Sinyalleri

• Knowledge-Centric Self-Improvement, Oracle icin skill/memory evolution'da "bilgi merkezli geri besleme" tasarimini watchlist'e aldiriyor. Kaynak: https://arxiv.org/abs/2607.19592
• DynamicRubric, evaluator ve policy'nin birlikte evrildigi degerlendirme dongusu oneriyor; trajectory scoring icin uygulanabilir. Kaynak: https://arxiv.org/abs/2607.20083
• Guardrails as Scapegoats, agent refusal'larinda guardrail kararinin faithful olup olmadigini audit etme ihtiyacini netlestiriyor. Kaynak: https://arxiv.org/abs/2607.19449

Kaynak Ozeti

• rss/news ailesi 1.410 raw item ve 273 unique item verdi; gunun guvenilir urun sinyalleri OpenAI Health, Hugging Face Nunchaku ve AWS agentic retrieval uzerinden geldi. Kaynak: https://openai.com/index/health-in-chatgpt ve https://huggingface.co/blog/nunchaku-diffusers ve https://aws.amazon.com/blogs/machine-learning/agentic-retrieval-for-amazon-bedrock-managed-knowledge-base/
• search ailesi 513 raw item ve 263 unique item verdi; Google News aggregator linkleri yine baskin, ama regulation/security/funding sinyalleri buradan geldi. Kaynak: https://arstechnica.com/tech-policy/2026/07/ai-kill-switch-act-would-let-trump-admin-order-shutdown-of-rogue-ai-systems/ ve https://techcrunch.com/2026/07/23/ai-chip-startup-etched-defies-skeptics-hits-10-3b-valuation-from-big-name-investors/
• academic/api ailesi 867 raw item ve 496 unique item verdi; arXiv cs.LG/stat.ML agirligi self-improvement, edge inference ve AI-science sinyallerini besledi. Kaynak: https://arxiv.org/abs/2607.19592 ve https://arxiv.org/abs/2607.19438 ve https://arxiv.org/abs/2607.19406
• community ailesi 70 raw item ve 57 unique item verdi; GitHub trending tarafinda gateway, architecture-as-code ve AI video editor sinyalleri one cikti. Kaynak: https://github.com/diegosouzapw/OmniRoute ve https://github.com/likec4/likec4 ve https://github.com/palmier-io/palmier-pro
• social ailesi 4.131 raw item ve 875 unique item verdi; X fallback agirligi yuksek oldugu icin bu kaynaklar ana kanit degil, tartisma sinyali olarak kullanildi. Kaynak: https://news.google.com/rss/articles/CBMiX0FVX3lxTE43ZDJZNFNwcmhYWnpQa0xJQVdheTJYRnp0WV9tU1hudEItWEtlb01tM0JQRkRtQkdhUy02eGsxWVZ0Z0k2YVVsWGV6SV9xVElQUkY2a18zY29qVXNiaWg4?oc=5

Coverage / Blind Spots

• 5/5 kaynak ailesi kapandi: rss/news, search, community, social, academic/api; missing family yok. Kaynak: https://openai.com/index/health-in-chatgpt
• Thin topic yok; top topics Launches, Regulation, Models, Security ve Infra etrafinda toplandi. Kaynak: https://techcrunch.com/2026/07/23/runway-bets-on-ai-model-routing-as-generative-media-gets-crowded/ ve https://arstechnica.com/tech-policy/2026/07/ai-kill-switch-act-would-let-trump-admin-order-shutdown-of-rogue-ai-systems/
• En buyuk kalite riski search/social tarafinda aggregator linkleri; canonical URL cozumleme hala iyilestirme adayi. Kaynak: https://news.google.com/rss/articles/CBMiaEFVX3lxTE5zd29CQVptVHlVM24ybFVEd1BCVGM5ZGtBdk1FOERyaEdrbDg5ZzZqOTVtRmF1bzlLWjRNRkhIWUpOY2lNeXAxV1R2cGZrYmc4TEgyTG1zLTh6MGlLbmw1V216OHRhaVd50gFuQVVfeXFMT0JqaTVPNEl3bnFZQjRnTXV1Nm9yTGhtZUdqTjBqWEtPa2ZVc1JIeFlPendXdWNGUzlDOW1xY3oycEJ3RkpnVkl4RnJtVmZncFA2MkJqeGdRZk9tb1VyZ19aR2cyNzFYTHY1ejRObEE?oc=5

Bu Gece Sistem Ne Ogrendi

• "Launches, Regulation, Models" ucgeni devam ediyor; bugun bunun yeni formu Health in ChatGPT, Runway router ve AI Kill Switch Act oldu. Kaynak: https://openai.com/index/health-in-chatgpt ve https://techcrunch.com/2026/07/23/runway-bets-on-ai-model-routing-as-generative-media-gets-crowded/ ve https://arstechnica.com/tech-policy/2026/07/ai-kill-switch-act-would-let-trump-admin-order-shutdown-of-rogue-ai-systems/
• Agent guvenligi artik iki ayri hatta izlenmeli: remediation agent'lari ve sandbox escape riskleri. Kaynak: https://news.google.com/rss/articles/CBMiaEFVX3lxTE5zd29CQVptVHlVM24ybFVEd1BCVGM5ZGtBdk1FOERyaEdrbDg5ZzZqOTVtRmF1bzlLWjRNRkhIWUpOY2lNeXAxV1R2cGZrYmc4TEgyTG1zLTh6MGlLbmw1V216OHRhaVd50gFuQVVfeXFMT0JqaTVPNEl3bnFZQjRnTXV1Nm9yTGhtZUdqTjBqWEtPa2ZVc1JIeFlPendXdWNGUzlDOW1xY3oycEJ3RkpnVkl4RnJtVmZncFA2MkJqeGdRZk9tb1VyZ19aR2cyNzFYTHY1ejRObEE?oc=5 ve https://news.google.com/rss/articles/CBMiggFBVV95cUxNM3FvNkRxdExRQ2JKaGM1VmRURUdUWF95OGVQZ1VhMFNmS1RoYzhYNTFZZ1RPTjJ1OU11SHFteVdISHNwRXdibUl1MVFBTmdxTEFHWkxNeVBERDk4TEdBMFNWemsta1psV29IeTU3N05lQWJ5VkxNZ1NwQ0JJNU9FMHRn?oc=5
• Edge/secure inference sinyali guclendi; BaseRT ve confidential GPU benchmark'lari local/private inference tasarimlarina dogrudan girdi saglar. Kaynak: https://arxiv.org/abs/2607.19438 ve https://arxiv.org/abs/2607.19353

Dedupe & Kalite Notu

Bu rapordaki ana maddeler onceki 3 gunun raporlarindaki tekrar basliklardan elendi. Merge: 5 dosya islendi, 7.708 giris -> 6.893 cikis, 815 tekrar atildi. Dedupe: 6.893 giris -> 2.332 yeni unique, 4.537 fingerprint tekrar, 24 semantik tekrar. Raporlanabilir item: 1.964; signal-only/blocked unique: 368.