Oracle Gece Arastirma v3 - 2026-07-31

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

Oracle Gece Arastirma v3 - 2026-07-31

Otomatik derleme | v3 pipeline | 7.178 raw item | 3.197 input unique | 2.874 raporlanabilir item | 108 benzersiz kaynak | 5/5 kaynak ailesi

Gunun Ozeti

Bugunun ana ekseni robotik, agent guvenligi ve AI altyapisinin maliyet/enerji sinirlarina dogru kaydi. Google DeepMind, Gemini Robotics ER 2 ile video understanding, task orchestration ve multi-robot collaboration hattini buyuttu; bu, AI agent anlatiminin fiziksel dunyaya tasindigini gosteriyor. Kaynak: https://deepmind.google/blog/gemini-robotics-er-2-powering-robotics-with-video-understanding-task-orchestration-and-multi-robot-collaboration/

Kurumsal hatta NVIDIA, AWS, GitHub, Snyk ve Oracle tarafindan gelen sinyaller agent deployment, MCP, prompt caching, coding assistant ve model menusu katmanlarinin artik production security problemi oldugunu gosterdi. Kaynak: https://developer.nvidia.com/blog/four-ways-to-deploy-more-secure-ai-agents/ ve https://github.blog/changelog/2026-07-29-copilot-code-review-agent-skills-and-mcp-now-generally-available ve https://snyk.io/blog/announcing-snyk-studio-integration-snowflake-cortex-code/

Altyapida Nscale-Anyscale satin almasi, AWS capex artisi, Samsung memory crunch uyarisi ve EU AI gigafactory basvurulari ayni tabloyu veriyor: AI rekabeti model benchmark'larindan compute, memory, enerji ve sovereign capacity planlamasina donuyor. Kaynak: https://techcrunch.com/2026/07/30/nscale-buys-anyscale-as-it-seeks-to-own-more-of-the-ai-compute-stack/ ve https://www.cnbc.com/2026/07/30/amazon-amzn-q2-earnings-report-2026.html ve https://www.theregister.com/systems/2026/07/30/samsung-warns-memory-crunch-will-last-through-2028-as-profit-rises-19-fold/5281142 ve https://sifted.eu/articles/eu-opens-applications-for-e10bn-ai-gigafactory-scheme/

Top 7

1. Google DeepMind, Gemini Robotics ER 2'yi duyurdu; model video understanding, task orchestration ve multi-robot collaboration ile humanoid/embodied agent hattini guclendiriyor. Kaynak: https://deepmind.google/blog/gemini-robotics-er-2-powering-robotics-with-video-understanding-task-orchestration-and-multi-robot-collaboration/
2. NVIDIA, secure AI agent deployment icin dort mimari yol yayinladi; agent rollout artik guardrail degil, identity, isolation, observability ve policy plane meselesi. Kaynak: https://developer.nvidia.com/blog/four-ways-to-deploy-more-secure-ai-agents/
3. AWS, Kimi K3'u SageMaker HyperPod ve EKS uzerinde calistirma rehberi yayinladi; open-weight frontier modellerin cloud deployment pattern'i netlesiyor. Kaynak: https://aws.amazon.com/blogs/machine-learning/deploying-kimi-k3-on-aws/
4. Google, AI yardimiyla Chrome'da iki surumde 1.072 security bug fixledigini soyledi; AI destekli bug hunting patch temposunu yukseltiyor. Kaynak: https://www.bleepingcomputer.com/news/google/google-says-ai-helped-chrome-fix-1-072-security-bugs-in-two-releases/
5. Okta, AI security startup Permiso'yu satin aldi; identity security tarafi agent/AI threat surface icin konsolide oluyor. Kaynak: https://techcrunch.com/2026/07/30/okta-buys-ai-security-startup-permiso-source-says-for-about-200m/
6. EU, 10 milyar euroluk AI gigafactory basvurularini acti; sovereign compute artik Avrupa'nin resmi sanayi politikasi. Kaynak: https://sifted.eu/articles/eu-opens-applications-for-e10bn-ai-gigafactory-scheme/
7. Nscale, Anyscale'i satin alarak AI compute stack'inde daha fazla katmani sahiplenmeye gidiyor; infra oyunculari model serving ve orchestration'a yukari tirmaniyor. Kaynak: https://techcrunch.com/2026/07/30/nscale-buys-anyscale-as-it-seeks-to-own-more-of-the-ai-compute-stack/

Trend Analizi

• Spike: Databricks, Gemini, Haiku, Robotics ve Runway bugunun hizli yukselen basliklari; ozellikle Gemini + Robotics birlikteliği DeepMind ER 2 duyurusuyla anlam kazandi. Kaynak: https://deepmind.google/blog/gemini-robotics-er-2-powering-robotics-with-video-understanding-task-orchestration-and-multi-robot-collaboration/
• Yukselis: AI Agents, AI Regulation, AI Safety, Amazon ve Anthropic trend penceresinde yukseliste; bugunku agent security, capex ve regulatory enforcement haberleri bu trendi dogruluyor. Kaynak: https://developer.nvidia.com/blog/four-ways-to-deploy-more-secure-ai-agents/ ve https://www.cnbc.com/2026/07/30/amazon-amzn-q2-earnings-report-2026.html
• Azalis: GPT-4.1 ve Llama 4 dususte; topluluk akisi daha cok Kimi K3, GLM-5.2, Gemma 4, K-EXAONE ve edge/local inference etrafinda donuyor. Kaynak: https://www.reddit.com/r/LocalLLaMA/comments/1vb5td8/running_glm52_on_tinybox/ ve https://www.reddit.com/r/LocalLLaMA/comments/1vazdxp/lg_ai_research_releases_kexaone_20_750b_a37b/

KAT-1 - Meta / Ekosistem

• Meta, AI ile yeni uygulama gelistirmenin hizlandigini ve daha fazla app gelecegini soyledi; buyuk platformlar AI'yi sadece assistant degil, product factory olarak konumluyor. Kaynak: https://techcrunch.com/2026/07/30/meta-says-ai-is-making-it-easier-to-build-new-apps-and-more-are-coming/
• Microsoft'un OpenAI ve Anthropic ile daha acik rekabete girdigi anlatimi guclendi; hyperscaler ortakliklari artik ayni anda hem kanal hem rakip. Kaynak: https://techcrunch.com/2026/07/29/microsoft-is-openly-competing-with-openai-anthropic-more-than-ever

KAT-2 - LLM & Model Guncellemeleri

• AWS, Bedrock uzerinde OpenAI GPT-5.6 modelleri icin explicit prompt caching duyurdu; model maliyeti artik sadece token fiyati degil, cache orchestration verimliligi. Kaynak: https://aws.amazon.com/blogs/machine-learning/introducing-explicit-prompt-caching-for-openai-gpt-5-6-models-on-amazon-bedrock/
• Hugging Face trending modellerinde Kimi K3, Unlimited-OCR, GLM-5.2, VibeVoice-ASR-BitNet ve Mage-VL one cikti; open-weight model akisi model pazarini surekli yeniliyor. Kaynak: https://huggingface.co/moonshotai/Kimi-K3 ve https://huggingface.co/baidu/Unlimited-OCR ve https://huggingface.co/zai-org/GLM-5.2

KAT-3 - Guvenlik

• Amazon, Debug ve Chalk npm supply-chain saldirilarini Kuzey Kore baglantili bir gruba bagladi; AI coding ve package automation arttikca dependency trust daha kritik. Kaynak: https://www.bleepingcomputer.com/news/security/amazon-links-debug-chalk-npm-supply-chain-attacks-to-north-korean-hackers/
• VMware uc kritik acigi kapatti; auth bypass ve VM escape sinyali cloud/virtualization tarafinda blast radius'un yuksek kaldigini gosteriyor. Kaynak: https://www.bleepingcomputer.com/news/security/vmware-fixes-three-critical-flaws-allowing-auth-bypass-vm-escapes/
• Russian hackers Exchange OWA zero-day'i uzun sureli mailbox access icin kullandi; identity ve mail perimeter hala dusuk maliyetli giris noktasi. Kaynak: https://www.bleepingcomputer.com/news/security/russian-hackers-exploit-exchange-owa-zero-day-for-long-term-mailbox-access/

KAT-4 - Arastirma & Paper'lar

• TurboVLA, RTX 4090 uzerinde 32 Hz ve 1 GB altinda VRAM ile real-time vision-language-action iddiasi getiriyor; robotics inference icin latency/footprint ciddiye biniyor. Kaynak: https://arxiv.org/abs/2607.27205
• AgentSnare, autonomous penetration agents'i geciktirme, yonlendirme ve etkisizlestirme fikrini arastiriyor; offensive-agent savunmasi ayri bir research alani oluyor. Kaynak: https://arxiv.org/abs/2607.26998
• Hearsay, goruntu verilmediginde medical VLM'lerin diagnosis uydurabildigini gosteriyor; abstention ve evidence gating klinik AI icin zorunlu. Kaynak: https://arxiv.org/abs/2607.26886

KAT-5 - Reguelasyon & Politika

• Bir hakim, Trump yonetiminin Anthropic icin kullandigi "supply chain risk" etiketine yeterli kanit sunamadigini belirtti; AI policy savaslari mahkeme delil standardina giriyor. Kaynak: https://techcrunch.com/2026/07/30/judge-says-trump-admin-still-lacks-evidence-for-anthropic-supply-chain-risk-label/
• FCC'nin yeni Cinli robot ve inverter yasaklari, AI buildout'u robotik ve enerji ekipmani tedarik zinciriyle birlikte ele aldigini gosteriyor. Kaynak: https://the-decoder.com/fcc-bans-new-chinese-robots-and-power-inverters-to-protect-us-ai-buildout-from-foreign-threats/
• Eyalet savcilari AI enforcement icin geleneksel consumer protection yasalarini kullaniyor; yeni AI kanunu beklemeden yaptirim hattı aciliyor. Kaynak: https://news.google.com/rss/articles/CBMi3gFBVV95cUxQSWhuUVc2UjZ0ZkI5dk9vUUR1WFV0THpSeDc5YmtIanVUU2RNRFFIVTZwOGFfb1Z0OWhlMlpBYXJpaVJ1U2ZwN2dyM3RlRHdGWXh6VFEtWDZjN3lFQ0hLUmQwbkFNM1U5bUNWTWJ6QnRpRWdUOVctUElUcTM2NVFvSUVsUk1zSk1xZXNGWktJNS1NNkpvT21KZ09uMXlPb2lyeXo3V3Rjc0ZBUmR3cDFTSmJzQTNONU1KUTRwazlRVXJnbkpRRFVWMmFBNTJxemFnTWxLY196bGk2Uy1WUWc?oc=5

KAT-6 - Araclar & Framework'ler

• GitHub Copilot code review tarafinda agent skills ve MCP genel erisime acildi; coding review artik repository icinde agent toolchain'e baglaniyor. Kaynak: https://github.blog/changelog/2026-07-29-copilot-code-review-agent-skills-and-mcp-now-generally-available
• GitHub stacked pull requests public preview'a gecti; AI destekli uzun feature akislari icin kucuk ve sirali PR disiplini daha kolaylasiyor. Kaynak: https://github.blog/changelog/2026-07-30-stacked-pull-requests-are-now-in-public-preview
• PyTorch, FBTriton infra icin upstream ingestion ve hierarchical validation notlarini paylasti; AI compiler/kernel altyapisi validation discipline gerektiriyor. Kaynak: https://pytorch.org/blog/fbtriton-infra-upstream-ingestion-hierarchical-validation-ideals-vs-realities/

KAT-7 - AI Agent'lar

• Snyk Studio, Snowflake Cortex Code entegrasyonunu duyurdu; code assistant'larin guvenlik kontrolu IDE disindan data platformlarina tasiniyor. Kaynak: https://snyk.io/blog/announcing-snyk-studio-integration-snowflake-cortex-code/
• Databricks, legacy SQL'i agentlarla yeniden yazmaya yonelen araciyle migration pazarina girdi; agent'lar enterprise modernization islerinde rol aliyor. Kaynak: http://www.bing.com/news/apiclick.aspx?ref=FexRss&aid=&tid=6a6bc29df053478ebd46873aa477ab4b&url=https%3a%2f%2fwww.infoworld.com%2farticle%2f4203369%2fnew-databricks-tool-uses-ai-agents-to-rewrite-legacy-sql-at-scale.html&c=12485844682035854684&mkt=tr-tr
• Oracle, agent menusu icin Google Gemini ekledigini duyurdu; enterprise AI platformlari model-agnostic routing'i standart ozellik haline getiriyor. Kaynak: https://www.theregister.com/ai-and-ml/2026/07/30/oracle-adds-google-gemini-to-the-agent-menu/5281331

KAT-8 - Altyapi & Veri

• Amazon, 2026 capex hedefini 220 milyar dolara cikardi ve memory cost baskisini isaret etti; AI buildout'un maliyeti bellek tarafindan da belirleniyor. Kaynak: https://www.cnbc.com/2026/07/30/amazon-amzn-q2-earnings-report-2026.html
• Samsung, memory crunch'in 2028'e kadar surebilecegini soyledi; GPU ve inference planlarinda HBM/DRAM tedariki kritik bottleneck. Kaynak: https://www.theregister.com/systems/2026/07/30/samsung-warns-memory-crunch-will-last-through-2028-as-profit-rises-19-fold/5281142
• The Register, storage-inspired memory tech ile GPU'larin multiple-TB seviyesine cikabilecegini aktardi; memory hierarchy innovation model serving ekonomisini degistirebilir. Kaynak: https://www.theregister.com/storage/2026/07/30/gpus-could-explode-to-multiple-tb-with-new-storage-inspired-memory-tech/5281363

KAT-9 - Topluluk & Tartismalar

• LocalLLaMA'da Kimi K3'un local kullanim icin 1.56 TB'den 594 GB'a kadar sikistirilmis surumleri tartisildi; open-weight modellerde "calistirabilir miyim" sorusu model kalitesi kadar onemli. Kaynak: https://www.reddit.com/r/LocalLLaMA/comments/1va6ot2/kimi_k3_for_local_use_156tb_594gb_compressed_and/
• LocalLLaMA'da Turbo-fieldfare'in Gemma 4 26B'yi Apple Silicon'da 2 GB RAM ile calistirma iddiasi one cikti; edge/local inference toplulugu agresif optimizasyon pesinde. Kaynak: https://www.reddit.com/r/LocalLLaMA/comments/1vasnys/turbofieldfare_opensource_engine_running_gemma_4/
• LinkedIn'in "seems like AI slop" butonu, platformlarda AI content temizligi icin yeni reporting primitive'i olusturuyor. Kaynak: https://techcrunch.com/2026/07/30/linkedin-adds-a-button-to-report-ai-generated-slop/

SECURITY

• MIT Technology Review, LLM'leri saldiriya acik bir temel zafiyet hattiyla anlatti; model guvenligi prompt-level patch'ten daha derin bir sorun. Kaynak: https://www.technologyreview.com/2026/07/30/1140927/a-fundamental-flaw-leaves-llms-vulnerable-to-attack/
• Kritik Ruflo acigi, exposed MCP bridge uzerinden AI agent hijack riskini gundeme getirdi; MCP bridge'leri internet-facing ise threat model degisir. Kaynak: https://news.google.com/rss/articles/CBMiwwFBVV95cUxOUlVRNE1Jb1VUZ3FuMUhvOFo1MU9mSVllQVZkZ0RTX3Y4czMwRTVVQXdhUnhhSFc3czJ2RF8wZWU1WUVQalZGSGZ6R3BsY01jVkRHVllVWWRqaXAtWTh6em5sd1c2UnF1dGNwZ1JvQWVuM3NkeDM4M0hkLUEyX2NqcTd1VHljNDFYZ0FwRDllTkpmelJxay1zWWJkVkowX2Zoa2ZkMnpvZ1FRdzlLMWZZeEhTckdaY3JtWEJxei0tOXNYVWc?oc=5
• Not In My Git Yard, commit ve release zamaninda backdoor yakalamaya odaklaniyor; AI coding pipeline'larinda pre-release integrity kontrolu guclenmeli. Kaynak: https://arxiv.org/abs/2607.26719

REGULATION

• EU AI gigafactory programi, Avrupa'nin compute kapasitesini politika araci olarak ele aldigini gosteriyor. Kaynak: https://sifted.eu/articles/eu-opens-applications-for-e10bn-ai-gigafactory-scheme/
• Senato komitesi yasli yetiskinlere karsi AI-enabled fraud'u durdurmaya odaklandi; scam automation consumer protection riskini buyutuyor. Kaynak: https://news.google.com/rss/articles/CBMijgFBVV95cUxOdmhHSjJoYi01RzVGVnBDajRPbUtpZktOQ1RZZ2tqNjh0ZnEyMkdSZWVDb0VIRHlZVC1WalNCRjJMSThRVFdDMUd4ZmJuT0lKTEtJOXg2d0dpbTB1ZE1ENWpRRTNtQjJLSUNsNW5rX01IYTBMNE1NUVQ5SGctdU9xU2UtbHBmV0pISWhuRUpB?oc=5
• "AI did it" savunmasinin hukukta gecersiz kalacagi vurgulandi; org seviyesinde accountability zinciri kurulmak zorunda. Kaynak: https://www.theregister.com/legal/2026/07/30/excuses-like-ai-did-it-dont-exist-in-the-eyes-of-the-law/5280767

AI-SCIENCE

• EvoPINN, physics-informed neural networks icin executable algorithm discovery'yi agentic hale getiriyor; bilimsel modelleme agent tasarimina aciliyor. Kaynak: https://arxiv.org/abs/2607.26490
• Rad-JEPA 3D, CT volume'leri icin self-supervised radiology representation hattini guclendiriyor; unlabeled medical imaging verisi daha verimli kullanilabilir. Kaynak: https://arxiv.org/abs/2607.26196
• SCALPEL, medical vision-language representation icin LLM-powered encoder learning oneriyor; klinik multimodal foundation model pipeline'i zenginlesiyor. Kaynak: https://arxiv.org/abs/2607.26885

INFRA

• Hugging Face blog, idle GPU'lari grounded aircraft benzetmesiyle anlatip GPU management'i operasyonel verimlilik problemi olarak ele aldi. Kaynak: https://huggingface.co/blog/Dharma-AI/gpu-management
• InferScale, personalized LLM serving icin GPU-native KV injection oneriyor; long-context ve personalization memory cost'u runtime seviyesinde optimize edilebilir. Kaynak: https://arxiv.org/abs/2607.27090
• Cloudflare, cdnjs'i Developer Platform'a tasima deneyimini paylasti; buyuk olcekli platform migration'da dogfooding + validation kritik. Kaynak: https://blog.cloudflare.com/cdnjs-dev-platform-migration/

SAFETY

• Forecasting Trajectory-Level Safety Risks, black-box multi-turn interaction'larda riskleri erken tahmin etmeye calisiyor; agent safety post-mortem degil pre-mortem olmalı. Kaynak: https://arxiv.org/abs/2607.26820
• On-Policy Distillation for LLM Safety, template-robust realignment icin routing yaklasimi oneriyor; safety tuning'in prompt sablonlarina kilitlenme riski azaltilmak isteniyor. Kaynak: https://arxiv.org/abs/2607.27081
• Even More Deception, mixed-motive LLM multi-agent sistemlerde objective misalignment'i inceliyor; multi-agent setup'larda deception riski tek modelden farkli. Kaynak: https://arxiv.org/abs/2607.26120

WATCHLIST

• AI agents conduct open-ended AI research paper'i, autonomous research agent'lar icin erken kanit sunuyor; gercek performans metrikleri ve durma kriterleri izlenmeli. Kaynak: https://arxiv.org/abs/2607.27191
• Two Calls Beat Five Agents, local LLM'lerde multi-agent pipeline'larin self-refinement'a karsi her zaman ustun olmadigini sorguluyor; agent sayisi kalite garantisi degil. Kaynak: https://arxiv.org/abs/2607.26922
• Scores Are Not Decisions, LLM agent'larda tool acquisition icin cost-aware stopping oneriyor; tool kullanimi skor degil karar ekonomisi problemi. Kaynak: https://arxiv.org/abs/2607.27083
• From Tokens to Watt-hours, modern GPU'larda LLM inference enerji tahmini icin analitik model oneriyor; enerji metriği raporlara girmeli. Kaynak: https://arxiv.org/abs/2607.26571

TRENDS

• Robotics trendi Gemini Robotics ER 2, TurboVLA, Speech2Grasp ve HumanCLAW ile ayni gun birikti; embodiment tarafinda model + benchmark + edge inference birlikte ilerliyor. Kaynak: https://deepmind.google/blog/gemini-robotics-er-2-powering-robotics-with-video-understanding-task-orchestration-and-multi-robot-collaboration/ ve https://arxiv.org/abs/2607.27205 ve https://arxiv.org/abs/2607.26567 ve https://arxiv.org/abs/2607.27180
• AI Security trendi MCP bridge, npm supply chain, VMware, OWA ve AI bug hunting basliklariyla guclu; agent yayilimi klasik AppSec'i yeniden onemli hale getiriyor. Kaynak: https://www.bleepingcomputer.com/news/security/amazon-links-debug-chalk-npm-supply-chain-attacks-to-north-korean-hackers/ ve https://www.bleepingcomputer.com/news/security/vmware-fixes-three-critical-flaws-allowing-auth-bypass-vm-escapes/
• Open-weight trendi Kimi K3, GLM-5.2, Gemma 4 ve LG K-EXAONE tartismalariyla suruyor; local inference toplulugu model sikistirma ve RAM optimizasyonuna odaklandi. Kaynak: https://www.reddit.com/r/LocalLLaMA/comments/1vb5td8/running_glm52_on_tinybox/ ve https://www.reddit.com/r/LocalLLaMA/comments/1vazdxp/lg_ai_research_releases_kexaone_20_750b_a37b/

CikCik Paketi

• Sam Altman'in fiyat/intelligence tradeoff anlatimi sosyal akista yayildi; fiyat indirimi model rekabetinde ana narratif olmaya devam ediyor. Kaynak: https://news.google.com/rss/articles/CBMiWEFVX3lxTE9vYl8xLW9WVW9KYzM4SWtXdFUydG9QQ3MwS2lpY1ZwRG9tS0hYLVVGU3hSQ0RhVnhFVWRPMjg3dWZYa25lZ1d5ZTBtVlpqTWRVcWUxX1RpdFI?oc=5
• Philipp Schmid ve Omar Sanseviero akisi Gemini Robotics ER 2'nin Live API ve embodied reasoning tarafini one cikardi; developer sosyal katmani robotik duyuruyu hizli yaydi. Kaynak: https://news.google.com/rss/articles/CBMiYEFVX3lxTE1tOFhENFF2bXA3ekhJYWNqVDMyeEZrWFI1VF9USDVhczdmY2s5bE5aLTdJdkdaaUtQbk9STXlnMElFUnZhSmczUFRwUmJLcVVPSHJUVHhUdWZDMXVsRHBNTg?oc=5 ve https://news.google.com/rss/articles/CBMiYkFVX3lxTE1xY0YxeXozLUxJZ3BhVHd5eXo5ajVMUTJodG9YWGhnb19ucUdXa2txQ3NIeDlNcTk0dmZuUTNqUmhFRzc2XzhlQXhNQk5ZRjRNY1VId3RUTWZ2Z0VGQU8xQ0JR?oc=5
• Francois Chollet, ARC-AGI harness kullanimi ve bottleneck konularini gundeme tasidi; eval kalitesi model iddialarindan daha onemli hale geliyor. Kaynak: https://news.google.com/rss/articles/CBMiXkFVX3lxTE1xcUVIRktDTlB4TlZBN1NZOUtTN3pmOUphTURMVUVhZnFUcTdEbHhPRGJadEpFNVNlVTlnYjB5LTY2UzdjNGRwZzFTcnd2Y0RmUFRXUFB5UU1sNGtyOXc?oc=5
• LocalLLaMA'da "America Needs An Open-Source AI Strategy" basligi buyudu; ABD-Cin open-weight rekabeti topluluk tartismasina indi. Kaynak: https://www.reddit.com/r/LocalLLaMA/comments/1vb332c/america_needs_an_opensource_ai_strategy_cnbc/
• Mastodon akisi, Claude Code'u research synthesis pipeline'a ceviren Hyperresearch tartismasini tasidi; agentic research tooling talebi yukseliyor. Kaynak: https://mastodon.social/@camilleroux/117009759388669391

Oracle Self-Improvement Sinyalleri

• Pipeline 5/5 kaynak ailesini kapsadi; missing family ve empty topic yok. Kaynak: raw coverage summary, run 2026-07-31.
• Search akisi hala Google News/Bing aggregator linklerine yogun yaslaniyor; canonical URL cozumleme sonraki kalite artisi icin en net is. Kaynak: learning artifact, run 2026-07-31.
• Social collector 3.997 item uretti ama Twitter fallback kaynaklari gurultulu; CikCik paketi icin source reliability skoru ve eski tarih filtreleri sertlestirilmeli. Kaynak: raw-sources/2026-07-31/social.json
• Dedupe semantik tekrar sayisi 30'da kaldi; fingerprint tekrar sayisi yuksek oldugu icin URL canonicalization ve social-fallback title cleaning kaliteyi artirir. Kaynak: dedupe-v3 output, run 2026-07-31.

Kaynak Ozeti

• Raw toplam: 7.178 item; merge sonrasi 6.551; dedupe sonrasi 3.197 input unique; raporlanabilir 2.874 item. Kaynak: raw-sources/2026-07-31/merged.json ve raw-sources/2026-07-31/unique.json
• Kaynak ailesi kapsami: rss/news 1.237 raw, search 521 raw, community 67 raw, social 3.997 raw, academic/api 681 raw. Kaynak: coverage summary, run 2026-07-31.
• En baskin unique source family: social 1.502 unique; akademik/api 700 unique; search 304 unique; rss/news 312 unique; community 56 unique. Kaynak: coverage summary, run 2026-07-31.
• Top topics: Launches 1.028, Regulation 856, Models 842, Security 384, Agents 357. Kaynak: coverage summary, run 2026-07-31.

Coverage / Blind Spots

• Eksik family: yok; 5/5 collector ailesi tamamlandi. Kaynak: coverage summary, run 2026-07-31.
• Bos topic: yok; topics_covered 10/10. Kaynak: coverage summary, run 2026-07-31.
• Ince family: community 67 raw ile en dusuk hacimli aile; Reddit 429 ve HN bosluklari sinyal kaybina yol acabilir. Kaynak: community collector output, run 2026-07-31.
• Dusuk guven notu: Kantan News ve Twitter fallback sinyalleri ana kaynak yerine sadece signal/provenance olarak kullanilmali. Kaynak: rss collector output ve social collector output, run 2026-07-31.

Bu Gece Sistem Ne Ogrendi

• Rising entities onceki artifact'ta Meta, AI Regulation, Google, AI Agents, Anthropic, Claude, OpenAI ve RAG idi; bugun Google/Robotics, AI Agents/Security ve Regulation yine veriyle dogrulandi. Kaynak: learning artifact ve https://deepmind.google/blog/gemini-robotics-er-2-powering-robotics-with-video-understanding-task-orchestration-and-multi-robot-collaboration/
• Recurring themes Launches, Regulation, Models, Agents ve Tooling; bugun bu temalar GitHub MCP GA, NVIDIA secure agents, EU gigafactory ve open-weight local model tartismalariyla tekrar etti. Kaynak: https://github.blog/changelog/2026-07-29-copilot-code-review-agent-skills-and-mcp-now-generally-available ve https://sifted.eu/articles/eu-opens-applications-for-e10bn-ai-gigafactory-scheme/
• Kalici ders: Search akisi aggregator linklerine dayaniyor; rapor kalitesini artirmak icin canonical URL resolver ve old-date social fallback filtresi bir sonraki run'a eklenmeli. Kaynak: learning artifact, run 2026-07-31.

Dedupe & Kalite Notu

Bu rapordaki maddeler onceki 3 gunun raporlarindan elendi/dedupe edildi; 2026-07-28, 2026-07-29 ve 2026-07-30 raporlarindaki ana haberler tekrar edilmedi. Merge: 5 dosya, 7.178 giris -> 6.551 cikis; 627 tekrar atildi. Dedupe: 6.551 giris -> 3.196 yeni; 3.323 fingerprint tekrar, 30 semantik tekrar, 1 guncelleme.