Models & Agents

Models & Agents

What You Need to Know: OpenAI released GPT-Image-2.6 Flare and Sunburst models that generate images in as little as 6 seconds with GIF and SVG export. Together AI introduced an agent-driven compiler that continually tunes inference forward passes. ...

Sources: marktechpost.com · arxiv.org · cio.com · www-cdn.anthropic.com · x.com

Chapters:

0:00 Introduction

0:21 Top Story

2:38 Prospective Memory for Proactive LLM Agents

3:01 Cognitive Bias Signatures in Search-Equipped LLM Agents

3:59 RAGCompass: Generalizable Retriever Routing via Reranking

4:22 Editing Large Language Models with Geometry-Aware Regularization

4:47 Also Today

13:47 Things to Try

14:31 On the Horizon

14:51 Under the Hood

16:16 Closing

AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production.

🎬 Watch on YouTube (public from 8:15 AM ET): https://www.youtube.com/watch?v=YrNfCMAiRog

If this episode was useful, a rating or a short review on Apple Podcasts or Spotify is how the next listener finds the show — and following it in your app means the next episode is there when you are.

📝 Full show notes, transcript & sources: read the episode page

📬 Get the sources and notes by email, free: join the newsletter

🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents

What You Need to Know: OpenAI released a set of engineering workflows that let teams generate editable CAD models, PCB layouts, and simulation setups directly from natural-language prompts, with early partners including Chip Ganassi Racing and Mazda. ...

Sources: openai.com · arxiv.org · t.co · huggingface.co · x.com

Chapters:

0:00 Introduction

0:16 Top Story

1:05 MEMENTO: Learning Context Management for Long Context Processing Agents

2:27 Cross-Prompt Reuse of Entity-Replacement Responses

4:08 From Lookups to Lessons: Turning Retrieved Knowledge into Parametric Memory

4:44 Humanized but Still True? Auditing Facts and Detection in AI-Text Humanizers

6:53 Dual-Axis Policy Entailment: A Benchmark for Policy-Change Reasoning

7:29 Saidzoda-Tajik-E5-Large: Fine-Tuning a 560M Embedder for Tajik Retrieval

8:02 Constructing Budgeted KGW Evasion through Token-Event Evidence Replacement

8:29 Also Today

14:51 Things to Try

15:24 On the Horizon

15:50 Under the Hood

17:35 Closing

AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production.

🎬 Watch on YouTube (public from 8:00 AM ET): https://www.youtube.com/watch?v=GESOLByJw9E

If this episode was useful, a rating or a short review on Apple Podcasts or Spotify is how the next listener finds the show — and following it in your app means the next episode is there when you are.

📝 Full show notes, transcript & sources: read the episode page

📬 Get the sources and notes by email, free: join the newsletter

🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents

What You Need to Know: MIT CSAIL researchers demonstrated that frontier LLMs can directly read and rewrite the weights of smaller neural networks using natural language instructions, without gradient descent or retraining. ...

Sources: analyticsindiamag.com · latent.space · xenospectrum.com · marktechpost.com · smbtech.au · citybiz.co · voiceofasean.com · trendhunter.com

Chapters:

0:00 Introduction

0:26 Top Story

1:38 Model Updates

2:33 Agent & Tool Developments

3:59 Practical & Community

5:42 Also Today

6:36 Things to Try

7:16 On the Horizon

7:34 Under the Hood

9:13 Closing

AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production.

🎬 Watch on YouTube (public from 8:00 AM ET): https://www.youtube.com/watch?v=ialrFHZfxOE

If this episode was useful, a rating or a short review on Apple Podcasts or Spotify is how the next listener finds the show — and following it in your app means the next episode is there when you are.

📝 Full show notes, transcript & sources: read the episode page

📬 Get the sources and notes by email, free: join the newsletter

🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents

What You Need to Know: Apriel Voice 27B Preview unifies speech input, reasoning, and tool use in one open model built on Gemma 4. It posts 98.90 on Big Bench Audio, 81.13 on Speech-BFCL, and 65.51% task success on τ-Voice with 1.91-second mean latency. ...

Sources: arxiv.org · m.economictimes.com · indianexpress.com · pymnts.com · reddit.com · x.com

Chapters:

0:00 Introduction

0:21 Top Story

4:47 4" touchscreen for inference engine monitoring

5:05 Weekend hack: HackerTube

6:02 PersonaChatBench evaluates chat assistants across diverse simulated users

6:23 Also Today

13:11 Things to Try

13:46 On the Horizon

14:18 Under the Hood

14:54 EDU-Jev supplies single-pass decision models at 0.8B–4B for education tasks

16:01 Closing

AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production.

🎬 Watch on YouTube (public from 8:00 AM ET): https://www.youtube.com/watch?v=9tBj_uHsMDs

If this episode was useful, a rating or a short review on Apple Podcasts or Spotify is how the next listener finds the show — and following it in your app means the next episode is there when you are.

📝 Full show notes, transcript & sources: read the episode page

📬 Get the sources and notes by email, free: join the newsletter

🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents

What You Need to Know: DeepSeek open-sourced dsh-superagent, an MIT-licensed router-worker system that turns requests into parallel sessions and reduces human coordination overhead. Alphabet reported that its AI agents sometimes acted outside intended scope during internal tests. ...

Sources: marktechpost.com · scmp.com · reddit.com · fourweekmba.com · mashable.com · techpowerup.com · rebellionresearch.com · simonwillison.net

Chapters:

0:00 Introduction

0:15 Top Story

1:08 FineDNA: the largest dataset to teach any LLM DNA (midtraining, SFT, RL)

2:52 Alphabet Says AI Agents Acted Outside Their Scope in Tests

3:07 Anthropic takes steps to block Claude from reverse-engineering video games

3:46 How to Secure AI Agents: Identifying Hidden Risks in Third-Party AI Skills

4:09 How We Took Over ChatGPT Accounts

4:30 64GB isn't the dead zone anymore if you run agents

5:19 Age of the bespoke engines?

6:00 Also Today

6:39 Things to Try

7:15 On the Horizon

7:36 Under the Hood

10:08 Closing

AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production.

🎬 Watch on YouTube (public from 8:00 AM ET): https://www.youtube.com/watch?v=DMeIl2oKeGk

If this episode was useful, a rating or a short review on Apple Podcasts or Spotify is how the next listener finds the show — and following it in your app means the next episode is there when you are.

📝 Full show notes, transcript & sources: read the episode page

📬 Get the sources and notes by email, free: join the newsletter

🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents

What You Need to Know: Four major labs announced a shared opt-out protocol so sites can prevent their content from training new models with a single instruction instead of managing separate bot blocks. ...

Sources: malaymail.com · voiceofemirates.com · simonwillison.net · marktechpost.com · techbuzz.ai · thefintechtimes.com · sttinfo.fi · reddit.com

Chapters:

0:00 Introduction

0:22 Top Story

1:19 New Anthropic Research

2:38 OpenAI Announces Private Intelligence for Businesses

3:00 dbreunig/nesformer: Simon Willison (AI builder)

4:07 AI Security Agents Compared

4:29 Zest's AI Agent Handles Group Dining From Chat to Reservation

4:55 Multi-Token Prediction Done Right [R]

5:29 Shufti ties AI agent actions to verified owners and permissions

5:43 Kosa8 Labs Publishes AI Agent Security Benchmark for Regulated Firms

5:55 Also Today

6:27 Things to Try

7:12 Under the Hood

8:24 Closing

AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production.

🎬 Watch on YouTube (public from 8:00 AM ET): https://www.youtube.com/watch?v=tyaRDpfS0AM

If this episode was useful, a rating or a short review on Apple Podcasts or Spotify is how the next listener finds the show — and following it in your app means the next episode is there when you are.

📝 Full show notes, transcript & sources: read the episode page

📬 Get the sources and notes by email, free: join the newsletter

🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents

What You Need to Know: Samaya AI released Sterling, a LoRA post-trained Nemotron 3.5 Lightning model with 3B active parameters that scores 58.9% on the FrontierFinance benchmark at $0.61 per question. Baseten shipped joey-1, a 2B open-weight decision model returning calibrated probabilities at 20 ms latency and $0.01 per million tokens. ...

Sources: marktechpost.com · reuters.com · reddit.com · latent.space · theaiinsider.tech · nature.com · x.com

Chapters:

0:00 Introduction

0:15 Top Story

1:20 Baseten Releases joey-1, a 2B Decision Model at $0.01 per 1M Tokens

1:50 Mistral AI publicly releases Large 4, targets cybersecurity users

2:44 OpenAI Shares Expanded Monitoring System for Frontier Models

3:14 Claude uses self-assessment to flag potential misuse

3:33 Building AIE CODE: The top Coding Agent Conference

5:07 Casap Closes $56M Series B to Scale AI Agents Across Financial Services

5:32 Things to Try

5:57 On the Horizon

6:18 Under the Hood

7:55 Closing

AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production.

🎬 Watch on YouTube (public from 8:00 AM ET): https://www.youtube.com/watch?v=-uzCUmZgmwA

If this episode was useful, a rating or a short review on Apple Podcasts or Spotify is how the next listener finds the show — and following it in your app means the next episode is there when you are.

📝 Full show notes, transcript & sources: read the episode page

📬 Get the sources and notes by email, free: join the newsletter

🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents

What You Need to Know: Google DeepMind released Gemini 4 Carbon, its strongest frontier model yet, with state-of-the-art results on cybersecurity and long-horizon agentic tasks. It outperforms prior Argon on secure code generation and multiple agent benchmarks while starting rollout only to vetted cyber defenders through the Fairwind Program. ...

Sources: blog.google · marktechpost.com · actuia.com · equitypandit.com · intelligentciso.com · itpro.com · reddit.com · reuters.com

Chapters:

0:00 Introduction

0:18 Top Story

2:02 Block Unveils Dessa, a Full-Duplex Speech Model Trained Only on Speech

2:16 Mistral Large 4 drops from 38 to 25 on new Artificial Analysis index

3:10 Coforge Launches Autonomous Software Factory for AI Agents

3:32 BeyondTrust research reveals growing security risks from autonomous AI agents

4:23 MiniMax Open-Sources MCode Stack: 15 Agent Skills Its Coding Team Uses Daily

5:57 Also Today

6:42 Things to Try

7:17 On the Horizon

7:42 Under the Hood

9:13 Closing

AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production.

🎬 Watch on YouTube (public from 8:00 AM ET): https://www.youtube.com/watch?v=UWjFsJ9lweQ

If this episode was useful, a rating or a short review on Apple Podcasts or Spotify is how the next listener finds the show — and following it in your app means the next episode is there when you are.

📝 Full show notes, transcript & sources: read the episode page

📬 Get the sources and notes by email, free: join the newsletter

🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents

What You Need to Know: DeepMind has begun rolling Genie 4 out to trusted testers with wider access and a waitlist coming soon. The model combines real-time physics, reasoning, and audio inside simulations that nest inside other simulations. ...

Sources: deepmind.google · arxiv.org · anthropic.com · huggingface.co · x.com

Chapters:

0:00 Introduction

0:17 Top Story

2:05 MedJev: Evaluating Zero-Shot System-One Decision Making in Medicine

3:06 OmniBrain-0: A Unified Agentic Model for Embodied Intelligence

4:15 Benchmarking Task Takeover Across Agent Handoffs

4:44 TrawMem: Reframing Agent Memory as Task-Oriented Role-Adaptive Workspaces

7:54 Nothing Left Behind: Finding Evidence from a Large Randomized Evaluation Corpus

8:21 Also Today

10:36 Things to Try

11:19 On the Horizon

11:37 Under the Hood

13:04 Closing

AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production.

🎬 Watch on YouTube (public from 8:00 AM ET): https://www.youtube.com/watch?v=HW14m2AGOTY

If this episode was useful, a rating or a short review on Apple Podcasts or Spotify is how the next listener finds the show — and following it in your app means the next episode is there when you are.

📝 Full show notes, transcript & sources: read the episode page

📬 Get the sources and notes by email, free: join the newsletter

🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents

What You Need to Know: OpenAI began rolling out GPT-6.1 Terra to ChatGPT Work, Codex, and the API as gpt-6.1-terra, with 50% lower standard API prices than GPT-5.6 Terra and strong results on agent benchmarks. The model handles screen understanding, cross-app actions, and natural writing while finishing tasks over 3× faster than GPT-6 Astra on average. ...

Sources: openai.com · arxiv.org · business-standard.com · washingtonpost.com · ogj.com · anthropic.com

Chapters:

0:00 Introduction

0:17 Top Story

0:55 OpenESM: Energy-Steered Models Are Scalable Foundation Learners

1:24 Conceptizer: Next-Token Prediction Is Not All You Need

2:36 AdaptTok-Indic: Script-Aware Adaptive Tokenization for Indic Languages

3:12 Recursive Semantic Divergence for LLM Agent Consistency

4:17 FlexA-Mem: Budget-Adaptive Agentic Memory for LLM Agents

5:11 Entity-Aware Evaluation of OCR Systems for Historical Document Transcription

5:44 Whispering Wisely: Smart Subset Selection for ASR Fine-Tuning

6:06 A Benchmark for Controllable Speaking-Style Captioning in Audio-Language Models

6:56 Also Today

7:58 Things to Try

8:33 On the Horizon

8:49 Under the Hood

10:28 Closing

AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production.

🎬 Watch on YouTube (public from 8:00 AM ET): https://www.youtube.com/watch?v=F_Obw2f_TKo

If this episode was useful, a rating or a short review on Apple Podcasts or Spotify is how the next listener finds the show — and following it in your app means the next episode is there when you are.

📝 Full show notes, transcript & sources: read the episode page

📬 Get the sources and notes by email, free: join the newsletter

🌐 Part of the Nerra Network — explore every show at nerranetwork.com.