Anthropic, a prominent AI developer, has released an upgraded version of its Claude Instant conversational AI model, Claude Instant 1.2.
The new model demonstrates significantly improved capabilities in areas like math, coding, reasoning and safety compared to the previous 1.1 version.
More Accurate Answers with Less Hallucination
Claude Instant 1.2 incorporates strengths from Anthropic’s latest flagship Claude 2 model, resulting in longer, better-structured responses. And achieved higher scores in math and coding tests during evaluations.
On the Codex coding evaluation, it scored 58.7%, a significant jump from 52.8% with the older 1.1 version. The model also reached 86.7% on the GSM8K grade-school math benchmark, up from 80.9%.
Anthropic says the upgraded Claude Instant exhibits increased safety through reduced hallucinations and resistance to manipulation attempts.
In internal red team evaluations designed to trick the AI, 1.2 proved the most secure model yet. This should provide businesses with more consistently accurate information.
Improved Performance Across Multiple Tasks
Besides math and coding, Claude Instant 1.2 demonstrates better summarisation, casual conversation, question answering and document comprehension abilities.
Although still slightly behind the original in some benchmarks, differences were minimal.
The new release follows formatting guidelines more precisely and generates lengthier, structured responses overall. It also shows gains in quote extraction, working with multiple languages, and drawing information from documents like PDFs.
Faster Processing, Lower Costs for Businesses
As a lightweight version of Claude 2, Claude Instant offers significantly faster processing and lower costs than Anthropic’s more advanced models. This makes it accessible for smaller companies and developers with limited budgets.
Businesses can now utilise the upgraded capabilities of 1.2 through Anthropic’s conversational AI API.
Pricing tiers are designed to meet different business needs and workloads.
The full Claude 2 model also remains available for those requiring maximum performance.
Anthropic states they have an exciting development roadmap planned for the Claude Instant line. The company aims to improve key metrics like safety and factual accuracy through slow, iterative enhancements.
They have also recently launched the beta website providing public access to Claude 2 Conversations. However, Claude Instant itself is currently only available through the API for business applications.
Also read:
A Leading Contender Among AI Assistants
Since OpenAI’s release of ChatGPT sparked intense interest in conversational AI, companies have raced to develop competing models.
Anthropic’s Claude line-up has emerged as a formidable rival, with capabilities approaching GPT-3.5 and GPT-4 in many areas.
Claude Instant 1.2 represents a significant step forward, strengthening accuracy and reasoning while reducing problematic hallucinations.
As Anthropic continues honing its models, Claude appears well-positioned to drive innovation in real-world AI applications. But only time will tell if it can maintain an edge against rapid progress from OpenAI and other tech giants.
