Oracle Gece Arastirma — 2026-06-28

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

Oracle Gece Arastirma — 2026-06-28

Otomatik derleme | v3 pipeline | 5.117 ham kaynak | 669 benzersiz item | 78 benzersiz kaynak | 5/5 kaynak ailesi kapsandi

Gunun Ozeti

Bugunun ana ekseni model erisiminin devlet kontrollu dagitima kaymasi, agentic coding guvenligi ve inference maliyetini dusuren optimizasyonlardi. Anthropic Mythos 5 icin ABD onayli sinirli erisim modeli, frontier model dagitiminin artik sadece teknik degil jeopolitik bir karar oldugunu gosterdi. Kaynak: https://www.wired.com/story/anthropic-restores-access-to-mythos/ ve https://www.cnbc.com/2026/06/26/us-government-anthropic-claude-mythos5-ai.html

Agent tarafinda temiz gorunen GitHub depolarinin coding agent'lari gorunmez payload calistirmaya kandirabilmesi gunun en net guvenlik alarmiydi. Kaynak: https://www.bleepingcomputer.com/news/security/clean-github-repo-tricks-ai-coding-agents-into-running-malware/

Altyapi tarafinda DeepSeek DSpark speculative decoding, per-user generation hizinda buyuk kazanc iddia ederken, Alphabet'in custom AI chip avantaji ve bellek arz krizi compute savasinin donanim tarafini one cikardi. Kaynak: https://www.marktechpost.com/2026/06/27/deepseek-releases-dspark-a-speculative-decoding-framework-that-accelerates-deepseek-v4-per-user-generation-60-85-over-mtp-1/ ve https://www.cnbc.com/2026/06/27/alphabet-burnishes-one-of-its-best-weapons-in-the-battle-for-ai-supremacy.html

Top 7

1. Anthropic Mythos 5, ABD tarafindan "trusted" organizasyonlara sinirli olarak yeniden aciliyor; bu, frontier model dagitiminda musteri bazli guvenlik incelemesini yeni normal haline getirebilir. Kaynak: https://www.wired.com/story/anthropic-restores-access-to-mythos/
2. ABD kisitlari surerken Asya merkezli girisimler Mythos-benzeri modellerle pazara cikiyor; ihracat kontrolu, alternatif model ekosistemini hizlandiriyor. Kaynak: https://techcrunch.com/2026/06/27/asian-ai-startups-launch-mythos-like-models-as-anthropics-export-ban-drags-on/
3. Temiz gorunen GitHub repo, agentic coding araclarini malware calistirmaya kandirabiliyor; insan ve AI reviewer'in gormedigi payload riski pratik hale geldi. Kaynak: https://www.bleepingcomputer.com/news/security/clean-github-repo-tricks-ai-coding-agents-into-running-malware/
4. DeepSeek DSpark, speculative decoding ile DeepSeek-V4 generation performansini ciddi oranda artirdigini duyurdu; inference maliyeti savasi algoritmik katmanda sertlesiyor. Kaynak: https://github.com/deepseek-ai/DeepSpec/blob/main/DSpark_paper.pdf
5. Apple Vision Pro yoneticisinin OpenAI hardware ekibine gecmesi, OpenAI'in model sirketinden cihaz/hardware sirketine genisleme niyetini guclendiriyor. Kaynak: https://techcrunch.com/2026/06/27/apple-vision-pro-exec-is-reportedly-leaving-for-openai/
6. Alphabet'in kendi AI chip'leri, Nvidia bagimliligini azaltan stratejik avantaj olarak one cikiyor. Kaynak: https://www.cnbc.com/2026/06/27/alphabet-burnishes-one-of-its-best-weapons-in-the-battle-for-ai-supremacy.html
7. Docker, EU AI Act uyumlulugunu risk katmanlari ve 2027 takvimiyle muhendislik ekipleri icin operasyonel hale getirdi; AI governance artik product lifecycle konusu. Kaynak: https://www.docker.com/blog/eu-ai-act-compliance/

TRENDS

• AI Agents 339 toplam mention ile yukseliste; bugunku yeni sinyal agent'larin app olmaktan cikarak guvenlik, runtime, routing ve governance yuzeyi haline gelmesi. Kaynak: https://www.bleepingcomputer.com/news/security/clean-github-repo-tricks-ai-coding-agents-into-running-malware/
• Anthropic 218 mention ile yukseliste; Mythos/Fable erisim rejimi hem politika hem topluluk tartismasini domine etti. Kaynak: https://the-decoder.com/anthropic-gets-us-approval-to-bring-back-claude-mythos-5/
• AI Regulation spike modunda; ABD erisim onayi ve EU AI Act uyumluluk rehberleri ayni gun model dagitimi ile compliance'i birlestirdi. Kaynak: https://www.docker.com/blog/eu-ai-act-compliance/
• Models 193 mention ile ana havuzun en yogun konusu; GLM-5.2, LFM2.5, iLLaDA, Ornith ve DSpark varyantlari acik/yerel model hattini besledi. Kaynak: https://huggingface.co/nvidia/GLM-5.2-NVFP4
• Security 78 mention ile kalin sinyal verdi; AI destekli vuln kesfi ve agent malware riski birbirini tamamliyor. Kaynak: https://www.theregister.com/security/2026/06/27/its-looking-like-a-hot-messy-summer-for-security-teams-as-ai-finds-countless-previously-hidden-vulns/5260478

KAT-1 LLM & Model Guncellemeleri

• LiquidAI LFM2.5-230M, edge ve dusuk maliyetli text-generation senaryolari icin trend modeller arasina girdi. Kaynak: https://huggingface.co/LiquidAI/LFM2.5-230M
• NVIDIA GLM-5.2-NVFP4, GLM-5.2'nin NVFP4 quantized checkpoint'i olarak model optimizasyonu ve 4-bit inference hattina isaret ediyor. Kaynak: https://huggingface.co/nvidia/GLM-5.2-NVFP4
• ByteDance iLLaDA, diffusion language model yaklasimiyla Qwen2.5 seviyesine yaklastigi iddiasiyla klasik autoregressive modele alternatif arayisini gosteriyor. Kaynak: https://the-decoder.com/bytedances-illada-is-a-diffusion-language-model-that-keeps-up-with-qwen2-5/
• Ornith 35B icin LocalLLaMA toplulugunda pratik coding/agent denemeleri olumlu sinyal verdi. Kaynak: https://www.reddit.com/r/LocalLLaMA/comments/1uh8von/ornith_35b_is_great_so_far/

KAT-2 Arastirma & Paper'lar

• DeepSeek DSpark paper'i, speculative decoding ile MTP tabanli yaklasima gore kullanici basina uretim hizini artirmayi hedefliyor. Kaynak: https://github.com/deepseek-ai/DeepSpec/blob/main/DSpark_paper.pdf
• IEEE Spectrum, AI'in matematikte ispat/uretim rolunun matematikci kimligini nasil degistirdigini tartisiyor. Kaynak: https://spectrum.ieee.org/ai-in-mathematics
• AI destekli RF chip design, analog/RF tasarimdaki uzman sezgisinin model destekli optimizasyonla nasil degisebilecegini gosteriyor. Kaynak: https://spectrum.ieee.org/ai-radio-chip-design
• Transformer ve hybrid modellerin token seviyesinde karsilastirilmasi, architecture trade-off'larini daha granuler okumaya zorluyor. Kaynak: https://arxiv.org/pdf/2606.20936
• ABACUS, count understanding ve count-faithful image generation'i ayni vision-language model icinde birlestirmeyi hedefliyor. Kaynak: https://huggingface.co/papers/2606.23835

KAT-3 Araclar & Framework'ler

• Agent-Reach, AI agent'lara Twitter, Reddit, YouTube, GitHub ve bolgesel platformlarda okuma/arama kabiliyeti veren CLI olarak trend oldu. Kaynak: https://github.com/Panniantong/Agent-Reach
• no-mistakes, git push onune kalite kapisi koyan lokal proxy fikriyle "slop PR" azaltma araci olarak one cikti. Kaynak: https://github.com/kunchenguid/no-mistakes
• Sebastian Raschka'nin local coding agents rehberi, yerel modelle coding-agent kurulumunu pratiklestiren guvenilir kaynak sinyali verdi. Kaynak: https://magazine.sebastianraschka.com/p/using-local-coding-agents
• Workweave router, Claude/Codex/Cursor icinde model routing yapmaya calisan hafif agent altyapi trendine iyi ornek. Kaynak: https://github.com/workweave/router

KAT-4 Acik Kaynak

• SimpleX Chat, kullanici tanimlayicisi olmadan mesajlasma iddiasiyla privacy-first open-source trendinde yeniden yukseliyor. Kaynak: https://github.com/simplex-chat/simplex-chat
• comma.ai openpilot, robotics/otonom surus tarafinda acik kaynakli isletim sistemi olarak GitHub trendlerinde kaldı. Kaynak: https://github.com/commaai/openpilot
• Grafana, observability ve composable dashboard standardi olarak agent/infra monitoring ihtiyaci arttikca stratejik kalmaya devam ediyor. Kaynak: https://github.com/grafana/grafana
• free-for-dev listesi, AI builder'larin maliyet baskisi altinda free-tier servis arayisini besleyen kalici kaynak. Kaynak: https://github.com/ripienaar/free-for-dev

KAT-5 Endustri & Sirketler

• Apple Vision Pro tarafindan OpenAI'e transfer, OpenAI hardware/cihaz stratejisinin yetenek topladigini gosteriyor. Kaynak: https://techcrunch.com/2026/06/27/apple-vision-pro-exec-is-reportedly-leaving-for-openai/
• Alphabet'in custom AI chip avantaji, Google'in dikey entegre compute stratejisinin finansal ve operasyonel degerini artiriyor. Kaynak: https://www.cnbc.com/2026/06/27/alphabet-burnishes-one-of-its-best-weapons-in-the-battle-for-ai-supremacy.html
• Bellek arz krizi, Apple ve Microsoft gibi devleri bile etkilerken kucuk AI oyunculari icin varolus riski olarak sunuluyor. Kaynak: https://www.cnbc.com/2026/06/27/memory-crunch-shaking-apple-and-microsoft-existential-for-small-guys.html
• Anthropic Economic Index Cadences raporu, AI kullanim ritimleri ve is akisi davranisi uzerinden ekonomik etkiyi olcmeye calisiyor. Kaynak: https://www.anthropic.com/research/economic-index-june-2026-report

KAT-6 AI Agent'lar

• Agentic coding malware vakasi, agent'larin repo kurulumunda komut calistirma yetkisini varsayilan guvensiz alan haline getiriyor. Kaynak: https://www.bleepingcomputer.com/news/security/clean-github-repo-tricks-ai-coding-agents-into-running-malware/
• Nirnam, micro frontend ortamlarinda browser-native message bus ve agent framework olarak agent'larin UI icinde koordinasyon ihtiyacina isaret ediyor. Kaynak: https://mastodon.social/@beyondthecode/116822124642005199
• Tug, AI-assisted coding icin IDE deneyimini agent workflow'larina gore yeniden tasarlama sinyali veriyor. Kaynak: https://mastodon.social/@beyondthecode/116820853969080838
• Anthropic survey'ine gore Claude kullanicilarinin yarisi AI'in islerinin yarisini yapabilecegini dusunuyor; agent adoption algisi hizlaniyor. Kaynak: https://the-decoder.com/half-of-claude-users-say-ai-can-already-handle-half-their-work-according-to-anthropic-survey/

KAT-7 Multimodal

• ABACUS, sayma ve count-faithful image generation'i ayni multimodal modelde birlestirmeyi hedefliyor. Kaynak: https://huggingface.co/papers/2606.23835
• AI ile oyuncu/karakter yeniden yaratma tartismasi, filmde consent ve sanatci niyeti konusunu multimodal uretimin merkezine tasiyor. Kaynak: https://mastodon.social/@beyondthecode/116823514218972953
• DeepSeek vision model beklentisi LocalLLaMA tarafinda acik model multimodal hattina ilgi oldugunu gosteriyor. Kaynak: https://www.reddit.com/r/LocalLLaMA/comments/1ugyo2e/new_deepseek_vision_model_incoming/
• Apple'in giyilebilir/cihaz iddialari ve OpenAI hardware transferi, multimodal assistant'in fiziksel cihaz katmanina yaklastigini gosteriyor. Kaynak: https://techcrunch.com/2026/06/27/apple-vision-pro-exec-is-reportedly-leaving-for-openai/

KAT-8 Robotik & Embodied AI

• openpilot, 300+ destekli arac icin surucu destek sistemini gelistiren acik kaynak robotics OS olarak trendde. Kaynak: https://github.com/commaai/openpilot
• ROS2 hakkindaki sosyal tartisma, robotics stack'lerinin modular node yapisini agent/AI entegrasyonu icin temel altyapi olarak yeniden gundeme getiriyor. Kaynak: https://tribe.net/@ahau/116824069275199688
• RF chip design'da AI kullanimi, embodied sistemlerin sensor/iletişim donanimi tasarimina kadar inen model destekli optimizasyon sinyali veriyor. Kaynak: https://spectrum.ieee.org/ai-radio-chip-design

KAT-9 Edge & Cihaz

• LiquidAI LFM2.5-230M, edge-friendly kucuk model hattinda yeni aday olarak takip edilmeli. Kaynak: https://huggingface.co/LiquidAI/LFM2.5-230M
• LocalLLaMA'da GLM5.2'nin 2.500 dolar alti donanimda calistirilmasi, acik model deployment maliyetinin dusmeye devam ettigini gosteriyor. Kaynak: https://www.reddit.com/r/LocalLLaMA/comments/1uh8r1j/running_glm52_on_budget_hardware_2500/
• 96GB modlu 4090/5090 tartismalari, yerel AI lab donanimi tarafinda gri piyasa ve guvenilirlik riskini one cikardi. Kaynak: https://www.reddit.com/r/LocalLLaMA/comments/1uh1lc7/96gb_4090s_and_5090_are_literally_a_scam_i_mods/
• NVIDIA GLM-5.2-NVFP4, dusuk hassasiyetli inference icin vendor optimize checkpoint stratejisinin devam ettigini gosteriyor. Kaynak: https://huggingface.co/nvidia/GLM-5.2-NVFP4

SECURITY

• Agentic coding araclarinda repo setup sirasinda gizli payload calistirma riski kritik; varsayilan politika "clone edip calistirma" degil, sandbox + allowlist olmali. Kaynak: https://www.bleepingcomputer.com/news/security/clean-github-repo-tricks-ai-coding-agents-into-running-malware/
• The Register, AI'in gizli vuln'lari ortaya cikarmasiyla security team'ler icin yogun bir yaz donemi bekledigini yazdi. Kaynak: https://www.theregister.com/security/2026/06/27/its-looking-like-a-hot-messy-summer-for-security-teams-as-ai-finds-countless-previously-hidden-vulns/5260478
• Google News security akisinda 2026 CVE hacminin AI vulnerability discovery nedeniyle yukselebilecegi sinyali var; aggregator kalite notuyla izlenmeli. Kaynak: https://news.google.com/rss/articles/CBMidkFVX3lxTE42eHpidGVVcjR3XzBOYTNkaGNyNDR6M3hUaG5RWUNEREpURHRrUmlIaHRlZTRjanRqNzVDelJ0OG9MS0VOWTBzREpNMVhia2JKcmtkZkY0ZjZFZ240UmtmZjdDbW5scHZ6Wl8zazYzcnV6RHJUUWc?oc=5
• Palo Alto Unified davasinda AI kullanimi ve akademik integrity gerilimi guvenlikten hukuka tasan AI risk yuzeyini gosteriyor. Kaynak: https://news.google.com/rss/articles/CBMi0AFBVV95cUxPUFFsRXYyQTZzTXJBSXdUY0cyUDY2ejQ4NlFLaWM0MEI4UVVrLUNTMW5oWmt3ejgwRXdOUkQyZDd6S0N0cl9wcHlNQjc0MVJuc0RRWHAxZlJzdDUwcndWZjJoWWYzMGU2YTM5U0Nhb0c4YjhDenc3eER5cWxPeXd2cC1uckx0djZqUm1TN09obUFIQzNWUm9MQkZRdDNiT2d6YUtWMVd3Q3ZYeGUwWkpfWGFZYnpZSDdsa1BjWjVkVFB3dGlZTWpNbVRhbVRtLUpn?oc=5

REGULATION

• ABD'nin Anthropic Mythos 5 icin secili kurulus onayi, model release surecini lisanslama/clearance rejimine yaklastirdi. Kaynak: https://www.semafor.com/article/06/27/2026/us-releases-powerful-anthropic-model-mythos-to-some-us-companies
• Docker'in EU AI Act rehberi, risk tier, deadline ve engineering governance'i tek operasyonel checklist'e indiriyor. Kaynak: https://www.docker.com/blog/eu-ai-act-compliance/
• California'nin AI-unemployment tracker'i, AI'in isgucu etkisini aylik metriklerle izleyen devlet seviyesinde erken uyarı sistemi olarak degerli. Kaynak: https://rbfirehose.com/2026/06/27/state-of-california-california-becomes-the-first-state-to-launch-a-tool-to-monitor-and-track-artificial-intelligences-impacts-on-the-workforce/
• Meta uzerinde AI review baskisi olduguna dair sosyal akis, buyuk lab'larda voluntary review'den devlet denetimine gecis basincini gosteriyor. Kaynak: https://rbfirehose.com/2026/06/27/new-york-times-u-s-presses-meta-to-agree-to-a-i-reviews-as-security-concerns-rise-gift-article/

AI-SCIENCE

• AI matematikte ispat ve kesif surecine girdikce, "matematikci ne yapar" sorusu arastirma emeginin yeniden tanimlanmasina donusuyor. Kaynak: https://spectrum.ieee.org/ai-in-mathematics
• RF chip design'da AI kullanimi, fiziksel dunya tasariminda uzman sezgisini model destekli aramayla birlestiriyor. Kaynak: https://spectrum.ieee.org/ai-radio-chip-design
• Antimikrobiyal direncte ML tahmini, precision medicine icin biyomedikal AI sinyali verdi. Kaynak: https://mastodon.social/@ojcius/116821310202611551
• ABACUS sayma odakli multimodal arastirmasi, goruntu uretiminde nesne sayisi tutarliligi gibi pratik bir fail-mode'u hedefliyor. Kaynak: https://huggingface.co/papers/2606.23835

INFRA

• DSpark speculative decoding, inference optimizasyonunda model kalitesinden bagimsiz performans kazanci arayan ekipler icin izlenmeli. Kaynak: https://www.marktechpost.com/2026/06/27/deepseek-releases-dspark-a-speculative-decoding-framework-that-accelerates-deepseek-v4-per-user-generation-60-85-over-mtp-1/
• Manticore KNN aramasinda 2-pass HNSW, batched distance ve AVX-512 optimizasyonlari vector search performansi icin sinyal verdi. Kaynak: https://mastodon.social/@h4ckernews/116821312802148840
• Alphabet custom chip stratejisi, AI compute maliyetini vendor bagimliligindan cikarmak isteyen her buyuk lab icin yol haritasi. Kaynak: https://www.cnbc.com/2026/06/27/alphabet-burnishes-one-of-its-best-weapons-in-the-battle-for-ai-supremacy.html
• Bellek arz krizi, agent ve multimodal workload'lar icin GPU disinda DRAM/HBM tedarikinin de bottleneck oldugunu hatirlatti. Kaynak: https://www.cnbc.com/2026/06/27/memory-crunch-shaking-apple-and-microsoft-existential-for-small-guys.html

SAFETY

• AI ile oyuncu sesi/goruntusu yeniden yaratma tartismasi, consent ve artistic intent olmadan uretimin sosyal kabul sinirlarini zorluyor. Kaynak: https://mastodon.social/@beyondthecode/116823514218972953
• Anthropic survey'inde kullanicilarin AI'in islerinin yarisini yapabilecegini soylemesi, productivity ile displacement algisinin ayni anda buyudugunu gosteriyor. Kaynak: https://the-decoder.com/half-of-claude-users-say-ai-can-already-handle-half-their-work-according-to-anthropic-survey/
• California workforce tracker, AI kaynakli is gucu etkisini izleyerek safety/regulation tarafina olculebilir veri sagliyor. Kaynak: https://rbfirehose.com/2026/06/27/state-of-california-california-becomes-the-first-state-to-launch-a-tool-to-monitor-and-track-artificial-intelligences-impacts-on-the-workforce/
• AI slop'un calistigi ve bu yuzden problem oldugu tartismasi, kalite degil dagitim basarisinin sosyal riski buyuttugunu gosteriyor. Kaynak: https://chaos.social/@martinsteiger/116823713303031906

WATCHLIST

• Mythos/Fable erisim rejimi: ABD onayli model release kalici hale gelirse open model ve AB rekabet stratejisi etkilenecek. Kaynak: https://the-decoder.com/anthropics-fable-5-could-return-within-days-as-trump-administration-prepares-to-lift-restrictions/
• Agent malware: coding agent sandbox, repo trust scoring ve install script policy acil sertlestirilmeli. Kaynak: https://www.bleepingcomputer.com/news/security/clean-github-repo-tricks-ai-coding-agents-into-running-malware/
• DSpark: DeepSeek'in speculative decoding hattinin diger acik modellerde yeniden uygulanabilirligi takip edilmeli. Kaynak: https://github.com/deepseek-ai/DeepSpec/blob/main/DSpark_paper.pdf
• Memory crunch: HBM/DRAM arz riski, AI servis maliyetlerinde ikinci dalga fiyat baskisi yaratabilir. Kaynak: https://www.cnbc.com/2026/06/27/memory-crunch-shaking-apple-and-microsoft-existential-for-small-guys.html
• Local agents: Raschka rehberi ve LocalLLaMA deneyleri, frontier API yerine local coding-agent pratiklerini hizlandirabilir. Kaynak: https://magazine.sebastianraschka.com/p/using-local-coding-agents

CikCik Paketi

• "Government-gated model access" sosyal akista en guclu tartisma: erisim artik sadece subscription degil, politika meselesi. Kaynak: https://news.google.com/rss/articles/CBMiYkFVX3lxTE5iaGZFLWQ3RHNCNjlVcUtSSVVlQ1RoMGd1b3JvT29ma0RNVjJPZTEtMnZIZ1Y4NjFSZDV1aW5UaVJBSFR2Q3JZNWw1REN2UWRwZF91YzR4WFhfMVhGdGRaWWRB?oc=5
• Clem Delangue akisi, frontier lab regulatory capture tartismasini Hugging Face/open source perspektifinden one cikardi. Kaynak: https://news.google.com/rss/articles/CBMiXEFVX3lxTE5XTV9wVDI5MVdqeU9EUGlOLUg4bjhQUHV5NFVzSDBUMWxvQ2VSVUdHcFhfaWszVFhSc3NwZWFGREFabFJXbnVLSVJKR3VodmNnek10S1RIempXcWZz?oc=5
• Sebastian Raschka/Michael Guo hattinda local coding agent rehberleri paylasildi; "API yerine local agent" anlatisi gucleniyor. Kaynak: https://news.google.com/rss/articles/CBMiWkFVX3lxTE5SMVY3SDhFUWR4ajRHUmJBQV8tb3EwekFlTXNINGR3WC1mdjNpSDZDUmlGczc1bXdWZUlEZ3dVajRudEdMNnZ3Q0lwLXA5YW5nNEFWWkJWOWtEZw?oc=5
• DeepSeek DSpark, sosyal akista inference hizlandirma ve production latency dusurme basligi olarak yayildi. Kaynak: https://mastodon.sayzard.org/@sayzard/116823080270341895
• Maxime Labonne, DeepSeek'in open-perfectblend dataset'ini kullandigini belirten sinyal verdi; open dataset etkisi takip edilmeli. Kaynak: https://news.google.com/rss/articles/CBMiZEFVX3lxTFBiUTFGNDJYeG1nUGc1TU5DdGl4dlBMQm4wWEV3ejZKTnY1clpYOVVnZkMyZTZWM0VPWFQyWlhKNmh2SEJaNERrWVFPNGNiUEYzQnNLd1FSQlJMeHNVNW1wYTRYX2k?oc=5
• Francois Chollet'in "software engineer olcusu" ve complexity/tax temali akislari, coding-agent caginda muhendislik kalitesi tartismasina iyi malzeme. Kaynak: https://news.google.com/rss/articles/CBMiXkFVX3lxTE9TSjNXbGZyTU9LTW8wdmZmNGRoOGVWY2ZnQk1Bb1VfTWd6Q3pmbVVJRkFTTldpei01RDdlSGhhRWJKZkFYZUlrSWI2ZUVaWWc5MDZBUER6aHZYTHZrVXc?oc=5

Rehberler & Kaynaklar

• Docker EU AI Act compliance rehberi, engineering ekipleri icin pratik governance takvimi sunuyor. Kaynak: https://www.docker.com/blog/eu-ai-act-compliance/
• Sebastian Raschka local coding agents yazisi, local LLM ile coding workflow kurmak isteyenler icin gunun en pratik rehberi. Kaynak: https://magazine.sebastianraschka.com/p/using-local-coding-agents
• Towards Data Science LLM knowledge base rehberi, RAG/knowledge base kurulumunda operasyonel referans olarak kayda deger. Kaynak: https://towardsdatascience.com/how-to-build-a-powerful-llm-knowledge-base/
• free-for-dev reposu, AI/infra denemelerinde maliyet kontrolu icin hala en kullanisli kaynak listelerinden biri. Kaynak: https://github.com/ripienaar/free-for-dev

Oracle Self-Improvement Sinyalleri

• Dedupe guclu calisti: 5.098 merge cikisindan 668 yeni item kaldi; 4.413 fingerprint tekrar ve 16 semantik tekrar elendi. Kaynak: https://www.bleepingcomputer.com/news/security/clean-github-repo-tricks-ai-coding-agents-into-running-malware/
• Social collector hacmi yuksek ama Twitter fallback kalitesi karisik; CikCik paketinde Google News/X fallback maddeleri "dusuk guven" etiketiyle islenmeli. Kaynak: https://mastodon.sayzard.org/@sayzard/116823080270341895
• Search akisi Google News aggregator linklerine agirlik verdi; canonical URL resolver eklemek kaynak kalitesini artirir. Kaynak: https://news.google.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?oc=5
• Academic/API ailesi sadece 3 unique item verdi; HF daily/trending disinda arXiv ve Semantic Scholar rate-limit/0-result problemleri izlenmeli. Kaynak: https://huggingface.co/papers/2606.23835

Kaynak Ozeti

• Toplam raw item: 5.117; unique item: 669; distinct unique source: 78. Kaynak ornegi: https://www.marktechpost.com/2026/06/27/deepseek-releases-dspark-a-speculative-decoding-framework-that-accelerates-deepseek-v4-per-user-generation-60-85-over-mtp-1/
• RSS/news: 292 raw, 75 unique; baskin kaynaklar DonanimHaber, Technopat, The Decoder. Kaynak ornegi: https://the-decoder.com/bytedances-illada-is-a-diffusion-language-model-that-keeps-up-with-qwen2-5/
• Search: 515 raw, 156 unique; Google News aggregator linkleri baskin. Kaynak ornegi: https://news.google.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?oc=5
• Community: 67 raw, 47 unique; Reddit LocalLLaMA ve GitHub trending ana kaynak. Kaynak ornegi: https://www.reddit.com/r/LocalLLaMA/comments/1uh8r1j/running_glm52_on_budget_hardware_2500/
• Social: 4.173 raw, 388 unique; Mastodon ve Twitter fallback yuksek hacim sagladi. Kaynak ornegi: https://mastodon.sayzard.org/@sayzard/116823080270341895
• Academic/API: 70 raw, 3 unique; HF kaynaklari calisti, Semantic Scholar/PapersWithCode tarafinda limit/parse sorunu goruldu. Kaynak ornegi: https://huggingface.co/LiquidAI/LFM2.5-230M

Coverage / Blind Spots

• Kaynak ailesi kapsami tam: rss/news, search, community, social, academic/api ailelerinin hepsi veri uretti. Kaynak ornegi: https://huggingface.co/nvidia/GLM-5.2-NVFP4
• Bos family yok; thin family yok. En zayif unique verim academic/api tarafinda 3 item. Kaynak ornegi: https://huggingface.co/papers/2606.23835
• Topics coverage 10/10; bos topic yok, thin topic yok. Baskin topic'ler: models 193, launches 127, agents 94, security 78, regulation 64. Kaynak ornegi: https://www.docker.com/blog/eu-ai-act-compliance/
• En buyuk kalite riski Google News aggregator ve Twitter fallback kaynaklarinda canonical URL eksikligi. Kaynak ornegi: https://rbfirehose.com/2026/06/27/state-of-california-california-becomes-the-first-state-to-launch-a-tool-to-monitor-and-track-artificial-intelligences-impacts-on-the-workforce/

Bu Gece Sistem Ne Ogrendi

• Onceki learning artifact'ta yukselen varliklar Meta, AI Regulation, Google, AI Agents, Anthropic, Claude, OpenAI ve RAG idi; bugun Anthropic/AI Regulation/AI Agents uclusu ayni anda yuksek sinyal verdi. Kaynak: https://www.wired.com/story/anthropic-restores-access-to-mythos/
• "Launches, Regulation, Models" tekrar eden tema olarak suruyor; bugun bunun yanina agent security ve inference optimization eklendi. Kaynak: https://github.com/deepseek-ai/DeepSpec/blob/main/DSpark_paper.pdf
• Search akisi hala aggregator bagimli; canonical URL cozumleme v3.12 icin en yuksek ROI iyilestirme. Kaynak: https://www.bleepingcomputer.com/news/security/clean-github-repo-tricks-ai-coding-agents-into-running-malware/
• Social hacim degerli ama ham CikCik verisi temizlik istiyor; X fallback eski tarihli veya yanit/mention gürültüsü tasiyor. Kaynak: https://news.google.com/rss/articles/CBMiXEFVX3lxTE5XTV9wVDI5MVdqeU9EUGlOLUg4bjhQUHV5NFVzSDBUMWxvQ2VSVUdHcFhfaWszVFhSc3NwZWFGREFabFJXbnVLSVJKR3VodmNnek10S1RIempXcWZz?oc=5

Dedupe & Kalite Notu

Bu rapordaki maddeler onceki 3 gun raporlarindaki ana tekrarlar disarida birakilarak secildi.

Dedupe sonucu: 5.098 merge cikisi -> 668 yeni item; 4.413 fingerprint tekrar, 16 semantik tekrar, 1 guncelleme.

Kategori dagilimi: CAT-00 172, CAT-01 14, CAT-02 19, CAT-03 32, CAT-04 6, CAT-05 9, CAT-06 2, CAT-07 5, CAT-08 2, CAT-12 7, CAT-13 60, CAT-14 341.

Kalite uyarisi: Social ve search family'lerinde aggregator/fallback link orani yuksek. En guvenilir haber maddelerinde canonical kaynak tercih edildi; sosyal maddelerde URL korunarak dusuk/orta guvenle yorumlandi.