Future Forward - 104th Edition - Last Week in AI - A Primer on CLIO

Welcome to the 104th Edition of Future Forward - the Emerging Tech & AI Newsletter!

This newsletter aims to help you stay up-to-date on the latest trends in emerging technologies and AI. Subscribe to the newsletter today and never miss a beat!

Subscribe to the newsletter here.

Each edition covers top AI news from last week and an AI-related topic - Primers/Tutorials/ How AI is being used.

Here's what you can expect in this issue of the Emerging Tech & AI Newsletter:

• A summary of the top AI news from the past week.
• A Primer on CLIO

AI News from Last Week

The field of AI is experiencing rapid and continuous progress in various areas. Review of the notable advancements and trends from the last week below

Big Tech in AI:

1. Google open sourced AI(Perch) to understand animal sounds.
2. GPT-5 is available in CoPilot AI assistant.
3. Google launched AI tutor mode for students.
4. Microsoft announced CLIO (Cognitive Loop via In Situ Optimization) that enables LLMs to develop their own thought patterns.
5. Google's coding tool Jules is out of beta.
6. Google Deep Mind announced Genie 3.
7. Google added new storybook feature to Gemini app.
8. Google introduced Kaggle Game Arena.
9. Google released Gemini 2.5 Deep Think.
10. Apple has reportedly formed an "Answers, Knowledge and Information" team.
11. Alexa+ might deliver ads to users.
12. Meta is aiming to offload $2B worth of data center assets to partners.

Funding & VC Landscape:

1. Chai discovery raised USD 70M.
2. Perplexity acquired invisible.
3. Mistral is reportedly looking to raise USD 1Bn.
4. Trust AI raised $6M.
5. Tavily raised $25M.
6. WiseBee raised $2.5M.
7. Clay nets $100M at $3B.
8. Lyric grabbed $43.5M.
9. Motive grabbed $150M.
10. Decart hits $3.1B valuation.
11. Legion Raised $38M.

Other AI news:

1. Open AI released long awaited GPT-5.
2. MIT's AI predicts protein location in any cell through a new AI system called PUPS.
3. xAI plans to add ads in Grok's responses.
4. Browser Company launched subscription for its AI browser Dia.
5. Truth Social launched Truth Search AI, powered by Perplexity.
6. MiniMax released Speed 2.5
7. Open AI offers ChatGPT for $1 for US agencies.
8. Anthropic added automated security reviews.
9. xAI's Imagine tool rolled out in early access.
10. Open AI unvieled its open source models - gpt-oss-120b and gpt-oss-20b
11. Anthropic release Claude Opus 4.1
12. Eleven Labs introduced Eleven Music.
13. Alibaba released its Flash series of Qwen-3-Coder and other models via APIs.
14. Shopify added new agent focussed features.
15. Alibaba released Qwen image.
16. Perplexity partnered with open table.
17. Anthropic revoked Open AI's access to its APIs.

Liked the news summary? Subscribe to the newsletter to keep getting updates every week.

A Primer on CLIO

As artificial intelligence (AI) becomes increasingly embedded in scientific discovery and problem-solving, researchers have sought ways to improve not just accuracy, but also the transparency and controllability of AI reasoning processes. An emerging paradigm, known as the Cognitive Loop via In Situ Optimization (CLIO), is redefining how AI systems reason, adapt, and interact with users—especially in high-stakes, knowledge-intensive domains like science and medicine.

What is a Cognitive Loop?

A cognitive loop refers to a closed, recurring process in which an AI (or even a human) continually generates thoughts, reflects upon them, adapts strategies, and then restarts the process using feedback from prior cycles to improve outcomes. This loop is central to iterative problem solving and learning.

In traditional AI models, much of this iterative reasoning is established during post-training, and patterns are rigid. These models often allow little to no real-time control by end-users, making it difficult for scientists and experts to steer, inspect, or correct the AI’s thought process.

In Situ Optimization Explained

In situ means “in place” or “on the spot.” In the context of AI, in situ optimization refers to an AI’s ability to refine its reasoning and strategies while operating—dynamically adapting its behavior in the actual environment where problems are presented, rather than relying solely on pre-defined rules learned during offline

This is crucial for scientific discovery, where every new problem may require unique approaches and where adaptability can mean the difference between breakthrough and dead end.

How CLIO Works

CLIO, or Cognitive Loop via In Situ Optimization, blends these concepts into a new approach for large language models (LLMs) and other AI agents. Here’s how it works:

• Self-Formulation of Reasoning: Instead of merely applying fixed, pre-trained strategies, the AI actively formulates possible approaches to each new problem in real time.
• Internal Self-Reflection: The model creates reflection loops as it reasons, examining its progress, generating alternative hypotheses, and evaluating its own uncertainty at each step.
• Adaptive Behavior: When the AI detects uncertainty or low self-confidence, it can change course, try new strategies, or prompt the user for input and corrections.
• User Steering and Transparency: Scientists can observe the AI’s reasoning path, see uncertainty levels, and even interject manually at any stage. This enables correction and guidance for more reliable results.
• No Additional Post-Training: Unlike systems that require elaborate retraining to adjust behaviors, CLIO achieves advanced reasoning and adaptability without any additional post-training after deployment.

Benefits Over Traditional Approaches

• Steerability: Users are no longer passive consumers—they can steer the AI’s reasoning interactively. This is especially valuable for complex, dynamic domains like drug discovery or biomedical
• Explainability and Transparency: CLIO’s mechanisms make it possible to trace, critique, and edit the AI’s beliefs and decisions at each step, a key requirement for trustworthy AI in scientific tasks.
• Increased Performance: According to experimental results, integrating CLIO with GPT-4.1 increased accuracy on biomedical reasoning tasks from 8.55% to 22.37%, representing a 161.64% relative improvement—even surpassing more specialized models in high-effort reasoning scenarios.
• Model-Agnostic Application: CLIO can be layered onto various language models, enhancing performance without being tied to a specific architecture or training regimen

Source - https://www.microsoft.com/en-us/research/blog/self-adaptive-reasoning-for-science/

Implications for Science and Beyond

CLIO signals a new era of controllable, transparent, and ongoing reasoning in AI. By allowing AIs to optimize how they think in situ—while being observed and guided—the approach promises to make scientific discovery with AI more rigorous, reliable, and collaborative

This doesn't just benefit science; any domain requiring robust and explainable problem-solving (like finance, engineering, or legal reasoning) stands to gain from CLIO’s cognitive loop dynamics.

Conclusion

The Cognitive Loop via In Situ Optimization represents a crucial advance towards truly adaptive, explainable, and collaborative AI. By weaving together iterative self-reflection, user-steerability, and in-place optimization, CLIO opens the door for AI to become a transparent partner capable of matching the evolving demands of scientific inquiry and dynamic problem solving

References:

https://www.arxiv.org/pdf/2508.02789

Disclaimer: The content on CLIO was generated using AI tools. Let us know in case of any gaps.

Thanks for reading. See you next week!

Let's explore the future of technology together!

Your Turn:

Did you like the content? Share with your network. Subscribe now to get updates directly in your mailbox.