Local AI
394 episodes · Page 3 of 17
#6344: Training Text Models to Know When Not to Edit
In a 2023 WMT Marathi task, an LLM post-editor modified over 90% of sentences but improved only around 20%.
#6341: Why Fast Speech Detectors Still Struggle with Onsets
Aalto and Genelec experiments reveal human ears can detect sound timing shifts down to half a millisecond at certain frequencies.
#6332: Why Extra Voice Hours Settle Bilingual Speech
Fine-tuning Parakeet TDT on fifteen hours of bilingual speech cut error rates from 13.7% to 4.9% by resolving contextual variation.
#6317: Choosing Model Precision, Quantization, and Format
Quantizing Qwen 2.5 9B to Q4_K_M boosted generation speed 2.7 times over BF16, while prompt processing rates stayed roughly unchanged.
#6315: Inside China's Mixed-Language AI Models
Domestic Chinese models typically draw sixty to eighty percent of their training data from Chinese, leaving massive room for English and code.
#6314: Why Forgotten Runpod Workloads Keep Billing
Runpod stops pods at a zero balance, but attached volume disks double in cost when stopped and unattached pods vanish entirely.
#6312: Choosing Between Modal, RunPod, and Vast.ai
RunPod lists an H100 SXM at $3.99 an hour on Secure Cloud, but its Community marketplace drops it to $2.69.
#6309: Why DeepMind Systems Live Outside the Gemini API
AlphaEvolve saved 0.7% of Google's compute, but running it requires an enterprise cloud environment to evaluate candidate code.
#6308: How Google DeepMind Absorbed the Gemini App
In October 2024, Google moved Sissie Hsiao's Gemini app team directly into DeepMind, turning a research lab into a consumer product division.
#6304: Why Semantic Desktop Search Struggles to See Files
Local vector indexing on macOS and Windows requires hours of idle, plugged-in time to parse drives, often with no progress bar.
#6303: Blind Testing AI Video Models for Architects
A matrix of sixteen prompts and two seeds across three video models produces ninety-six clips to generate, fund, and review.
#6302: Why Fine-Tuning Whisper Won't Stop Duplicated Text
Whisper evaluates transcripts against a gzip compression threshold of 2.4, rejecting repetitive text before retrying at higher temperatures.
#6300: Choosing the Right Audio Annotation Tool
Desktop tool ELAN supports independent timeline tiers to prevent overlapping sounds like car horns from flattening into speech transcripts.
#6299: Finding Recurring Urban Sounds With Four Examples
In DCASE evaluations, prototypical network baselines can locate specific sound events across long recordings using as few as five labeled samples.
#6298: Why Dictation Models Steal Other People's Voices
Stock speech models lack target-speaker conditioning, defaulting to transcribing whichever competing voice is loudest or most predictable.
#6297: Teaching Dictation Models to Ignore Other Voices
Evaluating Whisper on non-speech clips revealed entire invented phrases in roughly one percent of transcriptions.
#6288: What Happens Between Model Checkpoints
Extending reinforcement learning on Olmo 3 Think for 21 days yielded a 20-point leap on IFBench without starting fresh pretraining.
#6287: Why Model Names Conceal Changes Behind the API
One API identifier spanned ten distinct releases over nineteen months without changing the name submitted by the customer.
#6283: Why Terminal Benchmarks Miss Real Desktop Faults
A hardness audit revealed that over a third of universally failed benchmark tasks were broken rather than genuinely difficult.
#6278: Why Model Benchmarks Miss Small Workflow Friction
Switching Claude Code to Sonnet broke an everyday shell alias request that Opus solved simply by recognizing where to look in local context.
#6272: Teaching Audio Models to Hear Beyond Words
Speech recognition rewards ignoring background noise, but audio-language models train specifically on environmental sounds and temporal scenes.
#6265: Why Upgrading Embedding Models Demands Reindexing
Cohere's Embed Five Fast queries against a Pro index reach 98.4 percent of performance without regenerating stored document vectors.
#6264: Choosing Omni Embedding Models for Mixed Media
Jina's omni model trains only 0.35 percent of its weights while aligning with existing text indices.
#6255: Routing Claude Code Tasks to Local Ollama Models
A feature request to support local subagent endpoints in Claude Code was closed for inactivity after less than two months.