# \#llm **URL:** https://discuss.ai.google.dev/tag/llm/369.md [Latest](https://discuss.ai.google.dev/latest.md) · [Categories](https://discuss.ai.google.dev/categories.md) · [Tags](https://discuss.ai.google.dev/tags.md) --- ## [What Is Inside an LLM? Meet GlassBox](https://discuss.ai.google.dev/t/what-is-inside-an-llm-meet-glassbox/199320)
**Author:** [@Abdelhak\_Azzouz](https://discuss.ai.google.dev/u/Abdelhak_Azzouz)\ **Replies:** 0\ **Last updated:** [November 10, 2026, 1:09am UTC](https://discuss.ai.google.dev/t/what-is-inside-an-llm-meet-glassbox/199320 "2026-11-10T01:09:01Z")
When an AI model refuses a request or gets something wrong, its logs only show the text going in and out. They don’t show why it answered that way. Most LLM monitoring tools (LangSmith, OpenTelemetry, Weights & Biases)… --- ## [finishReason: RECITATION on users' OWN legal documents — also returns content:{} with no role, crashing strict SDK parsers](https://discuss.ai.google.dev/t/finishreason-recitation-on-users-own-legal-documents-also-returns-content-with-no-role-crashing-strict-sdk-parsers/171706)
**Author:** [@Wakkos](https://discuss.ai.google.dev/u/Wakkos)\ **Replies:** 2\ **Last updated:** [October 29, 2026, 7:28pm UTC](https://discuss.ai.google.dev/t/finishreason-recitation-on-users-own-legal-documents-also-returns-content-with-no-role-crashing-strict-sdk-parsers/171706 "2026-10-29T19:28:18Z")
Setup: Gemini Developer API (generativelanguage) via the Vercel AI SDK (@ai-sdk/google), streamText. Model: gemini-3-flash-preview. Use case: a legal assistant that summarizes/analyzes documents that our users upload the… --- ## [Request for manual review — app suspended for abuse](https://discuss.ai.google.dev/t/request-for-manual-review-app-suspended-for-abuse/190120)
**Author:** [@Henrik1234](https://discuss.ai.google.dev/u/Henrik1234)\ **Replies:** 4\ **Last updated:** [October 22, 2026, 1:24pm UTC](https://discuss.ai.google.dev/t/request-for-manual-review-app-suspended-for-abuse/190120 "2026-10-22T13:24:49Z")
Hello Google AI Studio team, My app is a Norwegian-language ferry route-planning tool. It retrieves weather information from multiple APIs and uses a static JSON file containing route information. The app now displays “… --- ## [Jules Ecosystem: A Jules-first autonomous software engineering platform across IDE, CLI, and Extension](https://discuss.ai.google.dev/t/jules-ecosystem-a-jules-first-autonomous-software-engineering-platform-across-ide-cli-and-extension/189925)
**Author:** [@Jack\_Sparow](https://discuss.ai.google.dev/u/Jack_Sparow)\ **Replies:** 0\ **Last updated:** [October 16, 2026, 8:47am UTC](https://discuss.ai.google.dev/t/jules-ecosystem-a-jules-first-autonomous-software-engineering-platform-across-ide-cli-and-extension/189925 "2026-10-16T08:47:51Z")
Hi everyone! I wanted to share an open-source project I’ve been building: Jules Ecosystem, a production-oriented autonomous software engineering platform designed to put Google Jules at the center of a controlled, multi-… --- ## [REPORT: SYSTEMIC CONTEXT BOUNDARY COLLAPSE](https://discuss.ai.google.dev/t/report-systemic-context-boundary-collapse/187454)
**Author:** [@Tyrone\_Steele](https://discuss.ai.google.dev/u/Tyrone_Steele)\ **Replies:** 0\ **Last updated:** [October 8, 2026, 3:05am UTC](https://discuss.ai.google.dev/t/report-systemic-context-boundary-collapse/187454 "2026-10-08T03:05:29Z")
TECHNICAL DILIGENCE REPORT: SYSTEMIC CONTEXT BOUNDARY COLLAPSE Date: October 7, 2026 Prepared for: Cornell / Technical Stakeholders Subject: Multi-Model Context-Weighting & Intent-Recognition Failures (Aura Benchmark R… --- ## [Clarification Request: Is a Plugin Sidecar Allowed to Use Antigravity Language Server GetModelResponse for PreToolUse Security Classification?](https://discuss.ai.google.dev/t/clarification-request-is-a-plugin-sidecar-allowed-to-use-antigravity-language-server-getmodelresponse-for-pretooluse-security-classification/186139)
**Author:** [@Windy\_Sin](https://discuss.ai.google.dev/u/Windy_Sin)\ **Replies:** 3\ **Last updated:** [October 3, 2026, 12:16pm UTC](https://discuss.ai.google.dev/t/clarification-request-is-a-plugin-sidecar-allowed-to-use-antigravity-language-server-getmodelresponse-for-pretooluse-security-classification/186139 "2026-10-03T12:16:57Z")
Clarification Request: Is a Plugin Sidecar Allowed to Use Antigravity Language Server GetModelResponse for PreToolUse Security Classification? I would like to clarify whether the following Antigravity plugin architecture… --- ## [Feedback: Creative Writing with Gemini 3.1 Issues and Improvements](https://discuss.ai.google.dev/t/feedback-creative-writing-with-gemini-3-1-issues-and-improvements/127438)
**Author:** [@Sammi\_W](https://discuss.ai.google.dev/u/Sammi_W)\ **Replies:** 7\ **Last updated:** [October 1, 2026, 2:37am UTC](https://discuss.ai.google.dev/t/feedback-creative-writing-with-gemini-3-1-issues-and-improvements/127438 "2026-10-01T02:37:33Z")
Feedback: 3.1 Regression in Long Form Collaborative Creative Writing I’ve been using Gemini models extensively since august '24(pre 2.0 days for sure) for long form collaborative creative writing and wanted to flag seve… --- ## [Optimal engineering approaches for integrating an AI model into a voice-command-only mapping application...](https://discuss.ai.google.dev/t/optimal-engineering-approaches-for-integrating-an-ai-model-into-a-voice-command-only-mapping-application/183506)
**Author:** [@Sami\_shniqat](https://discuss.ai.google.dev/u/Sami_shniqat)\ **Replies:** 0\ **Last updated:** [September 18, 2026, 3:55am UTC](https://discuss.ai.google.dev/t/optimal-engineering-approaches-for-integrating-an-ai-model-into-a-voice-command-only-mapping-application/183506 "2026-09-18T03:55:23Z")
Hello everyone, I am currently developing a local AI orchestration system for a smart mapping application named “Nashmi” within Android Studio, utilizing the llama.cpp library and offline GGUF models. I have successful… --- ## [🪰 Curing "As an AI language model..." and Cutting API Costs to Zero with Spiking Neural Networks](https://discuss.ai.google.dev/t/curing-as-an-ai-language-model-and-cutting-api-costs-to-zero-with-spiking-neural-networks/182649)
**Author:** [@Harry\_Hardon](https://discuss.ai.google.dev/u/Harry_Hardon)\ **Replies:** 0\ **Last updated:** [September 14, 2026, 2:18am UTC](https://discuss.ai.google.dev/t/curing-as-an-ai-language-model-and-cutting-api-costs-to-zero-with-spiking-neural-networks/182649 "2026-09-14T02:18:07Z")
Hello everyone, We are all familiar with the two great frustrations of building LLM agents: They constantly break character to remind you they are “just an AI.” They burn expensive API tokens to answer static, factual… --- ## [Notebook LM ignoring instructions](https://discuss.ai.google.dev/t/notebook-lm-ignoring-instructions/180683)
**Author:** [@Richard\_Sierra](https://discuss.ai.google.dev/u/Richard_Sierra)\ **Replies:** 2\ **Last updated:** [September 4, 2026, 8:37pm UTC](https://discuss.ai.google.dev/t/notebook-lm-ignoring-instructions/180683 "2026-09-04T20:37:43Z")
Hi, Notebook LM keeps ignoring my instructions. I have a comprehensive set of instructions in my protocol with verification and for some reason, they are being ignored. Any suggestions? --- ## [IDE spawns one language\_server 'multicall schedule' process per Scheduled Task and never reaps them on exit - orphans accumulate across sessions (504 processes / ~16 GB RAM)](https://discuss.ai.google.dev/t/ide-spawns-one-language-server-multicall-schedule-process-per-scheduled-task-and-never-reaps-them-on-exit-orphans-accumulate-across-sessions-504-processes-16-gb-ram/177949)
**Author:** [@Lukas\_Geiger](https://discuss.ai.google.dev/u/Lukas_Geiger)\ **Replies:** 2\ **Last updated:** [August 19, 2026, 7:54am UTC](https://discuss.ai.google.dev/t/ide-spawns-one-language-server-multicall-schedule-process-per-scheduled-task-and-never-reaps-them-on-exit-orphans-accumulate-across-sessions-504-processes-16-gb-ram/177949 "2026-08-19T07:54:16Z")
Summary The Antigravity IDE spawns one persistent language\_server.exe multicall schedule "\" agentapi new-conversation … child process per configured Scheduled Task at startup, but never terminates these scheduler p… --- ## [Gemini Pro no longer able to use NotebookLM notebooks as a source (normal consumer Pro account)](https://discuss.ai.google.dev/t/gemini-pro-no-longer-able-to-use-notebooklm-notebooks-as-a-source-normal-consumer-pro-account/136212)
**Author:** [@Marc\_Berger](https://discuss.ai.google.dev/u/Marc_Berger)\ **Replies:** 11\ **Last updated:** [August 13, 2026, 5:42am UTC](https://discuss.ai.google.dev/t/gemini-pro-no-longer-able-to-use-notebooklm-notebooks-as-a-source-normal-consumer-pro-account/136212 "2026-08-13T05:42:54Z")
I am using a regular consumer Google account with Gemini Advanced (Pro, family plan). Until around mid-March 2026, I was able to use my NotebookLM notebooks as a source in Gemini chats and in custom Gems. It was not alwa… --- ## [How do you record state for rerunnable multi-step LLM workflows?](https://discuss.ai.google.dev/t/how-do-you-record-state-for-rerunnable-multi-step-llm-workflows/175500)
**Author:** [@Dan-Levitan](https://discuss.ai.google.dev/u/Dan-Levitan)\ **Replies:** 0\ **Last updated:** [July 20, 2026, 12:31pm UTC](https://discuss.ai.google.dev/t/how-do-you-record-state-for-rerunnable-multi-step-llm-workflows/175500 "2026-07-20T12:31:23Z")
I am trying to understand how other developers record state once an LLM workflow becomes more than a single prompt. For one-off exploration, chat is fine. For repeatable work, the prompt text is only part of the system.… --- ## [Quantum Atomic RAG – Self-optimizing memory for LLMs via atomic shells](https://discuss.ai.google.dev/t/quantum-atomic-rag-self-optimizing-memory-for-llms-via-atomic-shells/175455)
**Author:** [@John\_Huerta](https://discuss.ai.google.dev/u/John_Huerta)\ **Replies:** 0\ **Last updated:** [July 20, 2026, 5:16am UTC](https://discuss.ai.google.dev/t/quantum-atomic-rag-self-optimizing-memory-for-llms-via-atomic-shells/175455 "2026-07-20T05:16:20Z")
I’ve been working on a project to address the “lost in the middle” phenomenon and the static nature of standard RAG systems. Most vector databases treat information as a flat, passive lookup table. I wanted something tha… --- ## [OpenAI and xAI Complete 1000 Turns, Gemini Repeatedly Returns HTTP 500 Errors](https://discuss.ai.google.dev/t/openai-and-xai-complete-1000-turns-gemini-repeatedly-returns-http-500-errors/173869)
**Author:** [@Szabolcs\_Krehlik](https://discuss.ai.google.dev/u/Szabolcs_Krehlik)\ **Replies:** 1\ **Last updated:** [July 8, 2026, 5:55am UTC](https://discuss.ai.google.dev/t/openai-and-xai-complete-1000-turns-gemini-repeatedly-returns-http-500-errors/173869 "2026-07-08T05:55:54Z")
Hello, I’m benchmarking multiple LLM providers using exactly the same production runtime, client implementation, and benchmark scenario. OpenAI 1000/1000 turns completed successfully. xAI Grok 1000/1000 turns comp… --- ## [Gemini Explicit Context Caching (cached\_content) drops system\_instruction in livekit.plugins.google — How to inject dynamic session variables?](https://discuss.ai.google.dev/t/gemini-explicit-context-caching-cached-content-drops-system-instruction-in-livekit-plugins-google-how-to-inject-dynamic-session-variables/170381)
**Author:** [@Harshita\_Sukumar\_Pat](https://discuss.ai.google.dev/u/Harshita_Sukumar_Pat)\ **Replies:** 0\ **Last updated:** [June 9, 2026, 7:05am UTC](https://discuss.ai.google.dev/t/gemini-explicit-context-caching-cached-content-drops-system-instruction-in-livekit-plugins-google-how-to-inject-dynamic-session-variables/170381 "2026-06-09T07:05:37Z")
We are working on optimizing costs for our voice agent utilizing gemini-2.5-flash via the livekit.plugins.google plugin. Our system prompts are large (5,000+ tokens), so we are leveraging Gemini’s Explicit Context Cachin… --- ## [The "Update-Break-Repeat" Cycle: A Call for Accountability to Premium Subscribers](https://discuss.ai.google.dev/t/the-update-break-repeat-cycle-a-call-for-accountability-to-premium-subscribers/137958)
**Author:** [@chunghiadaidong](https://discuss.ai.google.dev/u/chunghiadaidong)\ **Replies:** 2\ **Last updated:** [May 22, 2026, 2:57am UTC](https://discuss.ai.google.dev/t/the-update-break-repeat-cycle-a-call-for-accountability-to-premium-subscribers/137958 "2026-05-22T02:57:47Z")
I am writing this as a paying Google AI Ultra/Pro subscriber who, like many in the professional community, is reaching a breaking point. We are currently witnessing a frustratingly predictable pattern: Google releases an… --- ## [Deploying a PWA to a development environment. Main goal: convert it into an operational mobile app](https://discuss.ai.google.dev/t/deploying-a-pwa-to-a-development-environment-main-goal-convert-it-into-an-operational-mobile-app/109280)
**Author:** [@Ahmet\_Karayigit](https://discuss.ai.google.dev/u/Ahmet_Karayigit)\ **Replies:** 4\ **Last updated:** [May 1, 2026, 7:56pm UTC](https://discuss.ai.google.dev/t/deploying-a-pwa-to-a-development-environment-main-goal-convert-it-into-an-operational-mobile-app/109280 "2026-05-01T19:56:47Z")
I have been prototyping a PWA with Google AI Studio for a while. It is a big app with lots of features in it. I came to MVP level and I would like to add backend functions to make it operational. I don’t know if i need … --- ## [Agent Manager Completely Unresponsive - any LLM](https://discuss.ai.google.dev/t/agent-manager-completely-unresponsive-any-llm/137636)
**Author:** [@Ben\_Huffer](https://discuss.ai.google.dev/u/Ben_Huffer)\ **Replies:** 4\ **Last updated:** [April 24, 2026, 4:21am UTC](https://discuss.ai.google.dev/t/agent-manager-completely-unresponsive-any-llm/137636 "2026-04-24T04:21:20Z")
I’m doing a lot of internal debugging to no avail. The last few days Antigravity IDE has been completely unusable. The agent manager is completely dead. I’ve cleared cache. I’ve reset various features and things. I’ve… --- ## [Addressing Helpfulness Bias in Critical Contexts: The Zero Context Buffer Protocol. Dialogue for discussion](https://discuss.ai.google.dev/t/addressing-helpfulness-bias-in-critical-contexts-the-zero-context-buffer-protocol-dialogue-for-discussion/141989)
**Author:** [@Paskinz69](https://discuss.ai.google.dev/u/Paskinz69)\ **Replies:** 0\ **Last updated:** [April 23, 2026, 2:26pm UTC](https://discuss.ai.google.dev/t/addressing-helpfulness-bias-in-critical-contexts-the-zero-context-buffer-protocol-dialogue-for-discussion/141989 "2026-04-23T14:26:24Z")
Have this idea, tested and explained to model, text generate by model: Zero Context Buffer" Protocol Technical Validation (AI Architecture Review) Modern Large Language Models (LLMs) suffer from a critical vulnerabil… --- ## [Are there any age restrictions for using Google AI Studio?](https://discuss.ai.google.dev/t/are-there-any-age-restrictions-for-using-google-ai-studio/95249)
**Author:** [@dohunkm3](https://discuss.ai.google.dev/u/dohunkm3)\ **Replies:** 2\ **Last updated:** [April 20, 2026, 7:42am UTC](https://discuss.ai.google.dev/t/are-there-any-age-restrictions-for-using-google-ai-studio/95249 "2026-04-20T07:42:09Z")
I am interested in teaching elementary school students about large language models (LLMs) and having them practice using Google AI Studio. From my research, I have found that the Gemini API has age restrictions. I would … --- ## [Antigravity doesn't detect when a command is done](https://discuss.ai.google.dev/t/antigravity-doesnt-detect-when-a-command-is-done/128811)
**Author:** [@Damien\_Menigaux](https://discuss.ai.google.dev/u/Damien_Menigaux)\ **Replies:** 13\ **Last updated:** [April 17, 2026, 10:08am UTC](https://discuss.ai.google.dev/t/antigravity-doesnt-detect-when-a-command-is-done/128811 "2026-04-17T10:08:52Z")
Often times, I find that my LLM can’t properly detect that a command is done executing. The “Cancel” button remains visible as shown in the screenshot, so that means it’s probably the interface in itself, not the llm, ri… --- ## [LLM specialization field in API discovery?](https://discuss.ai.google.dev/t/llm-specialization-field-in-api-discovery/138719)
**Author:** [@Mikhail\_Kozyrev](https://discuss.ai.google.dev/u/Mikhail_Kozyrev)\ **Replies:** 0\ **Last updated:** [April 9, 2026, 1:03am UTC](https://discuss.ai.google.dev/t/llm-specialization-field-in-api-discovery/138719 "2026-04-09T01:03:16Z")
Hello, I’m making LLM neutral desktop AI agent. One of its features it’s picking a specific LLM for each task (or subtask inside one large task that was decomposed down). To do so agent uses several parameters and now… --- ## [This IDE is embarrassing](https://discuss.ai.google.dev/t/this-ide-is-embarrassing/138316)
**Author:** [@T\_K1](https://discuss.ai.google.dev/u/T_K1)\ **Replies:** 3\ **Last updated:** [April 8, 2026, 10:00pm UTC](https://discuss.ai.google.dev/t/this-ide-is-embarrassing/138316 "2026-04-08T22:00:24Z")
I don’t want to be mean, but I’d imagine google has the means to produce quality software. You’re not a startup, you have extraordinary ressources, yet the chat agent is full of bugs. Such as ‘the model output exceeded t… --- ## [Working Professionals No Longer Just Want to Chat](https://discuss.ai.google.dev/t/working-professionals-no-longer-just-want-to-chat/137867)
**Author:** [@One\_Only](https://discuss.ai.google.dev/u/One_Only)\ **Replies:** 2\ **Last updated:** [April 6, 2026, 7:05pm UTC](https://discuss.ai.google.dev/t/working-professionals-no-longer-just-want-to-chat/137867 "2026-04-06T19:05:44Z")
It seems the community has been quite heated lately regarding the quota issues in Google Antigravity. Honestly, it feels a bit ironic to complain about the reduced quota for the competitor’s Opus 4.6 model within Google… --- ## [Google Cloud Gemini LLM Batching running but never completing](https://discuss.ai.google.dev/t/google-cloud-gemini-llm-batching-running-but-never-completing/137404)
**Author:** [@OnQuestionAtaTime](https://discuss.ai.google.dev/u/OnQuestionAtaTime)\ **Replies:** 0\ **Last updated:** [April 1, 2026, 8:35am UTC](https://discuss.ai.google.dev/t/google-cloud-gemini-llm-batching-running-but-never-completing/137404 "2026-04-01T08:35:37Z")
For the last few days including today after the new google cloud terms of services, it’s impossible to get the batches sent through google cloud to finish out of the runnig states, for any preview model (3.1 or 2.5-sep-2… --- ## [I did not use the antigravity for the last 7 days, just opened it today](https://discuss.ai.google.dev/t/i-did-not-use-the-antigravity-for-the-last-7-days-just-opened-it-today/136289)
**Author:** [@Ramu\_M](https://discuss.ai.google.dev/u/Ramu_M)\ **Replies:** 1\ **Last updated:** [March 31, 2026, 4:49am UTC](https://discuss.ai.google.dev/t/i-did-not-use-the-antigravity-for-the-last-7-days-just-opened-it-today/136289 "2026-03-31T04:49:35Z")
Hi team, Does anyone face the issue below Claude LLM limits were exhausted when I hadn’t opened my IDE or AI studio for the last 7 days. I do not understand how that happens. I did not even use it; it seems like there… --- ## [Where should we turn to now?](https://discuss.ai.google.dev/t/where-should-we-turn-to-now/136672)
**Author:** [@cane\_a](https://discuss.ai.google.dev/u/cane_a)\ **Replies:** 5\ **Last updated:** [March 29, 2026, 3:22pm UTC](https://discuss.ai.google.dev/t/where-should-we-turn-to-now/136672 "2026-03-29T15:22:55Z")
Is Google providing a predictable service? Currently, the answer is a clear “No.” ​Should users simply sit back, wait for stabilization, and bear the financial loss on their own? I chose to subscribe to the paid tier ba… --- ## [The Gemini API is unreliable in the morning (PST)](https://discuss.ai.google.dev/t/the-gemini-api-is-unreliable-in-the-morning-pst/136607)
**Author:** [@higi](https://discuss.ai.google.dev/u/higi)\ **Replies:** 1\ **Last updated:** [March 28, 2026, 6:49pm UTC](https://discuss.ai.google.dev/t/the-gemini-api-is-unreliable-in-the-morning-pst/136607 "2026-03-28T18:49:57Z")
Hello, I are building a consumer facing product and rely on several LLM calls. I really like the intelligence/latency trade-off of Gemini under good conditions. However, every day at around 9am PST, the latency of the m… --- ## [NotebookLM API use case: Programmatic generation of Audio Overviews](https://discuss.ai.google.dev/t/notebooklm-api-use-case-programmatic-generation-of-audio-overviews/136322)
**Author:** [@sloaxleak](https://discuss.ai.google.dev/u/sloaxleak)\ **Replies:** 0\ **Last updated:** [March 27, 2026, 1:33am UTC](https://discuss.ai.google.dev/t/notebooklm-api-use-case-programmatic-generation-of-audio-overviews/136322 "2026-03-27T01:33:13Z")
Hi Google AI team, I’m building a platform that hosts scientific research papers. My use case is to programmatically generate 2-speaker “Audio Overviews” from PDFs as soon as they are uploaded to increase accessibility. … [Next page](https://discuss.ai.google.dev/tag/llm/369.md?match_all_tags=true&page=1&tags%5B%5D=llm)