Wow. And scary if and when proven, including on brains of which an eeg can't see, or neurodivergent brains negate, an action potential after simulation. EEG observations indicate air traffic controllers and surgeons have a propensity to not have a "normal" humans action potential after stimulation.
And, I'll wait for version 2 of VECTOR, especially as they named it VECTOR!
And await replication and higher benchmarks for "from OpenAI’s text-embeddings-3-small", and when the budget extends to the 6.5x cost of large. See "text-embedding-3-small: High-Quality Embeddings at Scale" below.
"We suggest our approach opens new avenues for understanding how humans dynamically organise and navigate conceptual knowledge in naturalistic settings." From "Charting trajectories of human thought using large language models" ... "Crucially, schema space trajectory properties predict language-independent behavioural markers, from fine-grained response times to trait-like communication patterns relevant to psychiatric assessment. The results demonstrate the power of aligning LLM representations with human cognition, enabling systematic study of conceptual organisation and thought through naturalistic language." ... "(here, we use OpenAI’s text-embeddings-3)" ... "from OpenAI’s text-embeddings-3-small). ... "We then generated a new 1536D semantic space embedding vector for each utterance using OpenAI’s text-embeddings-3-small model. ... https://arxiv.org/abs/2509.14455 ###
Follow the money... and model. MIRACL benchmark "text-embedding-3-small: High-Quality Embeddings at Scale By Jonathan PedoeemOct 31, 2025 ... "introduced a new economic reality for embeddings: - On multilingual retrieval tasks (MIRACL benchmark), performance jumped from 31.4% to 44.0%, a remarkable 13-point gain that makes the model significantly more effective for global applications - English task performance (MTEB benchmark) also improved from 61.0% to 62.3%, demonstrating consistent gains across languages
"But the real revolution comes in pricing. While ada-002 cost $0.10 per million tokens, text-embedding-3-small slashed this to just $0.02 per million tokens, an 80% reduction that fundamentally changes the calculus for embedding-heavy applications. This means a startup processing 100 million tokens daily would see their embedding costs drop from $10 to just $2, freeing up resources for other innovations. Better results for one-fifth the cost, it's a democratization of AI capabilities.
the Flexible Dimension Feature One of text-embedding-3-small's most innovative features is its implementation of Matryoshka Representation Learning, which allows developers to trim embedding dimensions on the fly without retraining the model. ... "The text-embedding-3 family consists of two models: text-embedding-3-small and text-embedding-3-large. Both models feature groundbreaking Matryoshka Representation Learning technology, which allows developers to adjust embedding dimensions dynamically without retraining the model. This flexibility enables optimization of storage costs and retrieval speed while maintaining semantic accuracy. Released in early 2024, these models have quickly become the industry standard for embedding generation due to their superior performance on benchmarks like MIRACL (multilingual retrieval) and MTEB (Massive Text Embedding Benchmark). The small model achieves 44.0% on MIRACL and offers a 5x cost reduction compared to ada-002, while the large model reaches 54.9% on MIRACL with richer 3072-dimensional representations. ... https://blog.promptlayer.com/text-embedding-3-small-high-quality-embeddings-at-scale/
Wow. And scary if and when proven, including on brains of which an eeg can't see, or neurodivergent brains negate, an action potential after simulation. EEG observations indicate air traffic controllers and surgeons have a propensity to not have a "normal" humans action potential after stimulation.
ReplyDeleteAnd, I'll wait for version 2 of VECTOR, especially as they named it VECTOR!
And await replication and higher benchmarks for "from OpenAI’s text-embeddings-3-small", and when the budget extends to the 6.5x cost of large. See "text-embedding-3-small: High-Quality Embeddings at Scale" below.
"We suggest our approach opens new avenues for understanding how humans dynamically organise and navigate conceptual knowledge in naturalistic settings." From "Charting trajectories of human thought using large language models"
...
"Crucially, schema space trajectory properties
predict language-independent behavioural markers, from fine-grained response times to trait-like
communication patterns relevant to psychiatric assessment. The results demonstrate the power of
aligning LLM representations with human cognition, enabling systematic study of conceptual
organisation and thought through naturalistic language."
...
"(here, we use OpenAI’s text-embeddings-3)"
...
"from OpenAI’s text-embeddings-3-small).
...
"We then generated a new 1536D semantic space embedding
vector for each utterance using OpenAI’s text-embeddings-3-small model.
...
https://arxiv.org/abs/2509.14455
###
Follow the money... and model.
MIRACL benchmark
"text-embedding-3-small: High-Quality Embeddings at Scale
By Jonathan PedoeemOct 31, 2025
...
"introduced a new economic reality for embeddings:
- On multilingual retrieval tasks (MIRACL benchmark), performance jumped from 31.4% to 44.0%, a remarkable 13-point gain that makes the model significantly more effective for global applications
- English task performance (MTEB benchmark) also improved from 61.0% to 62.3%, demonstrating consistent gains across languages
"But the real revolution comes in pricing. While ada-002 cost $0.10 per million tokens, text-embedding-3-small slashed this to just $0.02 per million tokens, an 80% reduction that fundamentally changes the calculus for embedding-heavy applications. This means a startup processing 100 million tokens daily would see their embedding costs drop from $10 to just $2, freeing up resources for other innovations. Better results for one-fifth the cost, it's a democratization of AI capabilities.
the Flexible Dimension Feature
One of text-embedding-3-small's most innovative features is its implementation of Matryoshka Representation Learning, which allows developers to trim embedding dimensions on the fly without retraining the model.
...
"The text-embedding-3 family consists of two models: text-embedding-3-small and text-embedding-3-large. Both models feature groundbreaking Matryoshka Representation Learning technology, which allows developers to adjust embedding dimensions dynamically without retraining the model. This flexibility enables optimization of storage costs and retrieval speed while maintaining semantic accuracy.
Released in early 2024, these models have quickly become the industry standard for embedding generation due to their superior performance on benchmarks like MIRACL (multilingual retrieval) and MTEB (Massive Text Embedding Benchmark). The small model achieves 44.0% on MIRACL and offers a 5x cost reduction compared to ada-002, while the large model reaches 54.9% on MIRACL with richer 3072-dimensional representations.
...
https://blog.promptlayer.com/text-embedding-3-small-high-quality-embeddings-at-scale/
SD