The continuing gap between the capabilities of Gemini Pro 3.1 (very good model) and the capabilities of the Gemini app/website is odd. The model can do what Claude/GPT can do, but there is a minimal harness for tools (file creation, research etc), no auditable CoT/actions, manual
LLMS
-
Improving Claude AI Transparency Through Published Tool Descriptions
By
–
The biggest challenge of using chat-based AI systems is that the details of what they can do are invisible – those tool descriptions are the missing manual, publishing them would be a huge benefit to people who want to get the most out of Claude
-
LLM Interface Evolution: Why Current Harnesses Fall Short
By
–
In more words: It feels like we’re due for a paradigm shift… we’re maxing out the utility of LLMs in their current harnesses (Codex, Claude Code). Every year or two, the capabilities of models improve such that the interfaces we use aren’t good enough to capture the full value
-
Model Evolution: From Forms to Agents and Beyond 2027
By
–
With how good the upcoming models are, I genuinely think “managing 14 Claude Code tabs” will look as stupid in 2027 as “filling out a GPT-3 form” looks now GPT-3 → forms
3.5 → chatbots
4 → workflows
5 → agents
next → ??? Every model leap outgrows the previous gen’s harness -

Position Encoding Evolution: Sinusoidal to RoPE to YaRN
By
–
Position Encoding Transformers
— Satya Mallick (@LearnOpenCV) 19 avril 2026
LLMs don't read words in order — they see everything at once.
Without position encoding, "the cat sat on the mat" and "the mat sat on the cat" are mathematically identical.
Full breakdown: sinusoidal → learned absolute → RoPE → YaRN →… pic.twitter.com/kyByXaJ1J1Position Encoding Transformers
LLMs don't read words in order — they see everything at once.
Without position encoding, "the cat sat on the mat" and "the mat sat on the cat" are mathematically identical.
Full breakdown: sinusoidal → learned absolute → RoPE → YaRN → -

Top AI Papers of the Week: April 13-19
By
–
The Top AI Papers of the Week (April 13 – 19) – AlphaEval
– AiScientist
– Auto-Diagnose
– Nemotron 3 Super
– Subliminal Learning
– Automated W2S Researcher
– Memory Transfer Learning Read on for more: -
Opus 4.6 Model Selector Tool Capabilities Change
By
–
I switched to Opus 4.6 in the model selector and asked about tools and seemed to get the same answer, which makes me think I would have needed to snapshot the tool descriptions a few weeks ago
-
Does 4.7 Model Tighten Hedging Language or Tool Use?
By
–
Did the 4.7 prompt tighten around hedging, or is it mostly tool-use phrasing?
-
How to start a conversation with Claude in learning mode
By
–
Empezar es muy sencillo:
— Nico (@nicos_ai) 19 avril 2026
Abre Claude en web o en la app, inicia una conversación y selecciona el Modo de Aprendizaje.
Haz una pregunta o sube un documento.
Claude se encarga del resto. pic.twitter.com/YeN74ZdXVuGetting started is very simple: Open Claude on web or in the app, start a conversation and select Learning Mode. Ask a question or upload a document. Claude takes care of the rest.
