BLT Stands for Belief, Locality, Transparency.
Belief Modeling, Local Knowledge, and Transparency are three things all today’s chatbots suck at, and here’s why this triple, much like Bacon, Lettuce, and Tomato in a sandwich, is a tasty extra.
1. Belief Modeling.
Contemporary chatbots largely fail at choosing the appropriate level of confidence when conveying uncertain information. Although there are mathematical ways to address that (e.g., belief functions), their comparative utility is yet to be explored.
A new generation of chatbots based on reasoning, or “thinking,” models feature test-time compute scaling. This means that before giving a response, the model produces intermediary output (“reasoning tokens”), expected to increase the model performance in proportion to the compute resources spent. This has not been the case with the previous generation such as ChatGPT-4o. Test-time compute scaling has a potential to improve Belief Modeling, although convincing evidence is lacking. OpenAI o1, Gemini Thinking, DeepSeek-R1, and Sky-T1, among others, are examples of reasoning models.
2. Local Knowledge.
Most chatbots rely on user data being sent to a cloud server for processing, which limits a casual user’s opportunity to rely on their own infrastructure and storage and compromises privacy.
Consider free-to-use, user-friendly tools such as Ollama and Open WebUI that allow one to try and run the best models locally on their own computer, including some local knowledge management.
3. Transparency Builds Trust.
Many chatbots have been launched, but they often lack comprehensive documentation and do not specify the models and approaches used.
Note that open source models, such as Llama, Qwen, DeepSeek, Mistral, Gemma, and others, offer a competitive experience while allowing for a more reproducible workflow when documentation is key.
BLT Or They’re The Same Chatbot.
In summary, consider not putting much trust in another AI startup for the time being, even if it promises a magic solution. Wait for the BLT features to start getting more traction.
In the meantime, commercial large language models (LLMs) such as GPT, Gemini, Claude, Grok, and a myriad of others are at your service, often free of charge or at an affordable cost.
Chat With The Team.
Be Learners Team! Embrace large language models and bookmark this page for future updates.