MWP – Artificial Intelligence Channel cover art

Artificial Intelligence

Episodes whose main subject is AI itself: building with, training, evaluating, or debating AI systems. AI used only as a tool for another topic goes to that topic.

1563 episodes Page 4 of 79 RSS Feed All feeds

#6493: Why Speech Models Forget Other Voices When Tuned on Yours

Adapting Whisper to one person's voice can trigger catastrophic forgetting, stripping away timestamps, segmentation, and accuracy on other speakers.

speech-recognitionai-evaluationfine-tuninglocal-aitraining-data

#6492: What Audio Models Learn Before Fine-Tuning

With just ten minutes of labeled audio on top of unlabeled pretraining, wav2vec 2.0 achieves a 4.8 word error rate on clean speech.

speech-recognitionai-evaluationlocal-aitext-to-speechfine-tuning

#6491: Agent Skills Break When Secrets Leave Your Laptop

Cloud environments deliberately strip uncommitted dot-env files during session uploads, leaving agent skills without working credentials.

ai-coding-assistantsmodel-context-protocolmulti-agent-systemsprompt-engineeringworkflow-automation

#6484: Why Narrow Speech Tasks Still Need Large Models

NVIDIA experiments show speech models trained from scratch on target domains had nearly double the error rate of fine-tuned pretrained models.

speech-recognitionlocal-aiai-inferenceai-evaluationfine-tuning

#6479: Why Research Agent Agreement Is Not Independence

When two distinct language models fail a task, they give the exact same wrong answer sixty percent of the time.

ai-evaluationreasoning-modelshallucinationsmulti-agent-systemsrag

#6475: USB Audio Limits on Android for Voice Models

Android caps native USB audio capture at 48 kilohertz, leaving higher rates entirely to apps packing their own custom drivers.

speech-recognitionmusicaudio-equipmentsmartphoneslocal-ai

#6473: Why Whispering Breaks Speech Recognition Physics

Whispering completely removes fundamental frequency and pitch, forcing speech recognition models to rely on subtle duration cues.

speech-recognitionphysicslocal-aitext-to-speechhallucinations

#6470: Why Computer Science Publishes in Conferences

When NeurIPS tested its review process with duplicate submissions, half the papers accepted by one committee were rejected by another.

taxonomy-and-classificationai-evaluationpodcastinghallucinationsopen-source

#6469: Why Switching Off One Attention Head Changes Everything

Masking just three decoder heads in Whisper drops hallucination on silent audio clips from 99.97% down to 24.1%.

transformer-architecturelocal-ai

#6468: Why Tiny Neural Gates Replaced Voice Activity Gates

Legacy WebRTC voice detection misses sixty-four percent of speech frames, while modern neural models compress full memory into a single megabyte.

speech-recognitionlocal-aiai-inferencetext-to-speechhvac

#6467: Voice Activity Detectors Need Adaptive Thresholds

When speech recognizers receive pure noise from a loose VAD gate, they confabulate plausible sentences rather than returning silence.

speech-recognitionhvaclocal-aitext-to-speechaudio-equipment

#6466: Conferences Rule Machine Learning, Not Journals

At NeurIPS 2025, over 21,000 papers required nearly 22,000 reviewers, yet fewer than half a percent earned oral presentations.

open-sourceai-evaluationtaxonomy-and-classificationtraining-data

#6465: Why Modern AI Models Cut Most Attention Layers

Frontier models like Kimi K3 now replace three out of four attention layers with delta-rule recurrence to curb massive KV caches.

context-and-memoryai-inferencelocal-aitransformer-architecturegpus

#6464: Teaching Speech Recognizers to Stay Quiet

Masking just three self-attention heads in Whisper large-v3 cuts hallucinations on non-speech audio by over eighty percent.

speech-recognitionlocal-aiai-evaluationai-inferencetraining-data

#6463: Why Training Speech Recognition From Scratch Fails

Thirty-two hours of speech provides just three hundred thousand words of text, making from-scratch speech models struggle with basic spelling.

speech-recognitionlocal-aiai-inferencetraining-datafine-tuning

#6462: When Adding a Reranker Hurts Search

NVIDIA benchmarked a cross-encoder upgrade that scored three points below its baseline, while other models surged fifteen.

ai-inferenceai-evaluationcontext-and-memoryrag

#6458: Why Synthetic Fine-Tuning Data Needs Tooling

Rewriting high-quality data synthetically gives a 3.4-point gain, but running the same process on low-quality data yields just 0.5.

ai-evaluationprompt-engineeringtraining-datafine-tuninghallucinations

#6457: Where to Learn LLM Training Without Generic Courses

A GPT-2 scale model that took OpenAI 168 hours to train in 2019 can now be replicated on a leaderboard in under two hours.

open-sourcetraining-datalocal-aifine-tuninggpus

#6456: The Unsung Builders of Modern Speech AI

OpenAI's Whisper lists famous co-founders among its six authors, yet shipped with no launch event, demo day, or canonical interview.

speech-recognitionlocal-aitext-to-speechtraining-dataopen-source

#6455: Turning Open-Source Triage into Subscription Credit

No software vendor currently applies peer-reviewed contribution credits directly against a customer's subscription renewal invoice.

open-sourceworkflow-automationai-coding-assistants