Agentic AI

[Paul Brown]

The next level of ai is here. "AI is leaving us all behind"

Google search

Conversation with Gemini​

is the next ai advancement 'agentic ai'


Yes, agentic AI is widely considered the defining advancement of the current era (2025–2026). While previous iterations focused on "Generative AI" (creating text or images based on prompts), the shift to "Agentic AI" marks a move toward systems that don't just talk, but act.


Here is a breakdown of why this is considered the "next big leap" and what it actually looks like in practice.

---

What Makes it "Agentic"?​

The core difference lies in autonomy. Traditional AI is passive; it waits for a prompt and provides a single response. Agentic AI is active; you give it a goal, and it determines the steps to reach it.

Feature	Generative AI (Chatbots)	Agentic AI (Agents)
Input	Specific prompt (e.g., "Write an email")	High-level goal (e.g., "Organize a trip")
Action	Generates text/media	Interacts with software, APIs, and tools
Logic	Linear response	Loops of reasoning, planning, and self-correction
Autonomy	Requires constant human steering	Can run in the background to complete tasks

---

Key Advancements in 2026​

In 2026, we are seeing this play out across several major sectors:

• Software Orchestration: Instead of you copy-pasting data between Excel and a CRM, an agent can "surf" your computer or use APIs to move data, send emails, and update records across different platforms.
• Self-Healing Systems: In cybersecurity and IT, agentic AI is being used to monitor networks and autonomously patch vulnerabilities the moment they are detected.
  
  
  
• Professional Services: In industries like law or finance, agents are being deployed to conduct multi-step research, verify findings across multiple databases, and draft comprehensive reports with minimal human oversight.
  
  
  
• The "Virtual Staff" Model: Major tech players (Microsoft, Salesforce, Google) have shifted their "Copilots" to "Agents" that act as delegated staff members rather than just writing assistants.

---

Challenges and "The Hype"​

While the potential is massive, there are significant hurdles:

• Cost & Compute: Running an agent that "thinks" through 50 steps to solve a problem uses significantly more processing power (and money) than a single chat response.
• Reliability: Agents can sometimes "hallucinate" an action (like deleting a file they shouldn't) or get stuck in logic loops.
• Governance: Many companies are struggling with how to give an AI "permission" to spend money or sign documents on their behalf.

The Consensus: We are currently in the "Year of the Agent." While the tech is still maturing and faces high project cancellation rates due to costs, it has fundamentally changed the goal of AI from "helping us write" to "doing the work for us."

 

[Paul Brown]

Video

Google Search

 

[Paul Brown]

Arena AI: The Official AI Ranking & LLM Leaderboard is an interesting concept. Use battle mode or one of the other modes. I use direct mode on a sons advice (drop down menu). Pitching one ai against another and evaluating the response.

 

[Paul Brown]

K-NN (K-Nearest Neighbors) and XGBRanker are two distinct machine learning approaches to horse racing. K-NN is an instance-based "lazy" algorithm that predicts an outcome by comparing a horse's profile to its historical K-most similar counterparts. XGBRanker is an eager, tree-based algorithm optimized specifically to order a group of horses (e.g., \(1^{\text{st}}, 2^{\text{nd}}, 3^{\text{rd}}\)) by learning how input features like odds, speed, and track conditions impact relative placement. [1, 2, 3, 4, 5]
Here is how they stack up against each other for handicapping:

K-Nearest Neighbors (K-NN)

• How it works: Finds the k closest past horse performances to the target horse based on metrics like distance, weight, and sectional times. It then averages the historical outcomes (e.g., final speed) to make a prediction. [1, 2]
• Strengths: Intuitive, requires no formal mathematical training phase, and is excellent for imputing missing data (e.g., estimating a horse's time over a new distance based on similar runners). [1, 2]
• Weaknesses: Struggles with high-dimensional data (analyzing too many attributes at once), computationally slow during inference, and cannot easily isolate complex interactions between factors. [1, 2]

XGBRanker (XGBoost Learning to Rank)

• How it works: Uses ensemble decision trees to evaluate all horses in a specific race at the same time and ranks them based on their relative likelihood of winning or placing. [1, 2]
• Strengths: State-of-the-art accuracy. It seamlessly handles large, noisy datasets and directly optimizes for ranking quality metrics such as NDCG or MAP. [1, 2, 3, 4]
• Weaknesses: Requires large volumes of historical race data to train effectively and is highly prone to overfitting if not tuned properly with regularization. [1]

The Bottom Line
K-NN works best as a localized, transparent tool for finding specific analogs (e.g., "Show me the last 5 times a 4-year-old ran heavy ground at this distance"). XGBRanker is vastly superior for building a comprehensive handicapping model, as it weighs complex, multi-dimensional data to predict the hierarchy of the entire field. [1, 2, 3, 4]

I can provide more technical details. Would you like me to: No thanks you've given me a headache.

 

[Paul Brown]

Expected Output for Your Race (Rough Estimate)With proper training, the model will likely rank:

1. DIAMOND RAIN — Strongest class + course history
2. DREAMASAR — Big improver
3. BOTAGOZ or SKY SAFARI...

---

Pro Tips for Better Performance

• Use more features (especially interaction and temporal ones).
• Try objective='rank:ndcg' instead of pairwise.
• Add query groups properly during training.
• Use early stopping to prevent overfitting.
• Blend with LightGBM Ranker for even better results.

Would you like me to:

• Expand the feature list with more realistic values for all 7 horses?
• Add probability calibration (turn rank scores into win %)?
• Show how to save/load the model and make daily predictions?

Just tell me what you need next!

 

[Chesham]

Here is the comprehensive tactical breakdown and data analysis for the 18:53 Newcastle, the Jenningsbet In Delves Hoppings Fillies' Stakes (Group 3). Run over 1m 2f on the artificial surface, this race attracts high-class fillies, often dropping back from top-tier turf targets.
Key Historical Trends & Preps
The Hoppings Stakes has a very distinct historical profile regarding where the winners come from:
The Elite Turf Drop-Downs: The premier prep routes for this AW Group 3 are high-class turf races. The Epsom Oaks (4 winners), Nottinghamshire Oaks (3 winners), and Middleton Stakes (2 winners) are the most potent feeders. Fillies dropping slightly in class from those elite turf battles thrive here.
The AW Transition: Form on synthetic surfaces is a plus, but raw turf class usually overrides AW-specific course form in this particular Group 3.
Pace Forecast & IPS Tactical Breakdown
The official Timeform Pace Forecast is Very Weak. In a small field of 7 on the all-weather, a crawl early on turns the race into a 3-furlong sprint, which severely punishes horses caught out the back.
Running Style Horses (Key Contenders) Tactical Outlook
Front Runners (IPS 1) Ashariba The Tactical Beneficiary. With an EPF1 of 0.576, she is the sole front-runner. In a Very Weak pace scenario, she will get an entirely uncontested lead and can dictate the fractions to suit herself.
Prominent / Mid (IPS 2/3) Alla Stella, Diamond Rain, Dreamasar The Tactical Sweet Spot. They will sit just off the slow pace. The Timeform Specific Pace Hint explicitly notes that the pace scenario shouldn't bother Diamond Rain. She has the tactical speed to hold her position.
Towards Rear (IPS 4/5) Sky Safari, Botagoz, Uluru The Tactical Victims. Trying to make up ground from the rear off a slow pace on the Newcastle tapeta is extremely difficult. They will be relying on their raw turn of foot to overcome the mathematical disadvantage.
Comprehensive Summary Chart
Below is the summary chart detailing the last 3 Timeform Ratings (TFR) and Timeform Speed Figures (Tfig) for today's runners.
Horse Odds PR% Last 3 TFR Last 3 Tfig
Diamond Rain 1.81 55.2% 109, 115, 116 21, ---, ---
Sky Safari 6.0 16.7% 103+, 104+, 108+ 70, 86, 87
Botagoz 6.8 14.7% 96+, 94, 82+ 96, 88, 68
Dreamasar 14.5 6.9% 110, 99+, 84 103, 74, 91
Alla Stella 19.0 5.3% 79?, 103+, 66 ---, 77, 55
Ashariba 32.0 3.1% 93+, 87, 98+ ---, ---, ---
Uluru 44.0 2.3% 82+, 85?, 88 ---, ---, 48
The Main Contenders: Data Profiles & Comment Analysis
1. Diamond Rain (Odds: 1.81 - Fav)
Timeform Interpretation: "Respectable third of 4 in a Group 2 at York in May when yard wasn't firing on all cylinders and this looks an ideal opportunity for her to resume winning ways."
The Verdict: The absolute class of the field (TFR 123). She won this exact race last year before placing in Group 1s. The Timeform pace hint explicitly clears her of any tactical concerns regarding the slow pace. She is the most likely winner by a significant margin.
2. Dreamasar (IRE) (Odds: 14.5)
Timeform Interpretation: "Progressive filly who has won both starts since stepping up in trip this year, including a listed race at Haydock... won her previous all-weather outing and this company may not prove beyond her given her upward trajectory."
The Verdict: The dark horse of the race. She carries a 'p' (117p) denoting further improvement and has won her last two starts stylishly. She maps to sit in mid-division and avoid the worst of the pace trap. At double-figure odds, she represents excellent each-way value.
3. Sky Safari (Odds: 6.0)
Timeform Interpretation: "Smart sort with 6 wins from 7 starts on speed-favouring all-weather tracks... Not seen to best effect in Group 2 on turf last time and will be suited by the switch back to today's surface."
The Verdict: A phenomenal all-weather specialist, but the data flashes a warning. She is a hold-up horse (IPS 4/5) heading into a "Very Weak" pace forecast. She will likely be out of position when the sprint for home begins.
4. Botagoz (IRE) (Odds: 6.8)
Timeform Interpretation: "Shaped well when third of 7 in a listed race at York in May, running on late. Progress required upped in grade but she's crying out for this longer trip and looks a likely improver."
The Verdict: A very highly-regarded 3-year-old (TFR 109p) taking on older mares. The step up to 1m 2f is a massive positive on pedigree, but like Sky Safari, her tendency to race off the pace could leave her with too much to do.
5. Ashariba (FR) (Odds: 32.0)
Timeform Interpretation: "Returned after 7 months off with respectable second of 9 in a similar event... Stable in red-hot form but this looks a significantly stronger race than she's contested previously."
The Verdict: The tactical wildcard. She is vastly outclassed on raw ratings, but she is the only front-runner in a race with no pace. If her red-hot stable has her primed, she could steal a place from the front at massive odds while the closers get stuck in a tactical mess.
Final Tactical Summary & Staking Plan
This Group 3 is dominated by the imposing presence of last year's winner, Diamond Rain, and the severe tactical implications of a "Very Weak" pace forecast.
Primary Win: Diamond Rain (1.81). She is a Group 1-calibre mare dropping into a Group 3. She won this race last year, has the tactical speed to handle the slow pace, and sets a standard the others cannot mathematically match.
The Value / Exacta Anchor: Dreamasar (14.5). A rapidly progressive filly who maps well in the pace setup. She is the most likely to capitalize if any of the market leaders misfire.
The Tactical Exotic: Ashariba (32.0). While outclassed on ratings, laying out a small fraction for her to hit the frame based purely on getting an uncontested lead in a slow-paced race is a mathematically sound longshot play.

 

[Chesham]

Sorry the above is cramped and not how it usually looks. Something not right with the Forum, as Attachment with screen shot do not appear and if editing the The Bar which has Font sizes, Bold and italics is greyed out.

 

[Paul Brown]

Sorry the above is cramped and not how it usually looks. Something not right with the Forum, as Attachment with screen shot do not appear and if editing the The Bar which has Font sizes, Bold and italics is greyed out.

Thanks. Thats looks fine.

 

[Chesham]

The Big Race Trend Hunter scores completely polarize this field, splitting it into a definitive "Big Two" historically, while introducing a fascinating conflict between raw trend alignment and tactical pace mapping.
Trend Hunter Impact: The Hoppings Stakes
The V1 Legacy scores reveal a massive gulf between the top two fillies and the rest of the field.
1. Diamond Rain (253 Points) — The Ultimate Dual-Threat
She already possessed the highest Timeform Rating (123) and the tactical clearance to handle the "Very Weak" pace. Now, she backs it up by dominating the historical trends. She matches a staggering 14 of the 24 key metrics, including crucial strike rates for her trainer/jockey combination and historical form in Group company. She is the complete package.
2. Sky Safari (250 Points) — The Profile vs. Pace Dilemma
She jumps off the page with a massive 250-point score, matching 15 different trends including "Highest OR" and elite historical win percentages in this class/code.
The Conflict: She has the historical DNA to win this, but the pace map is a nightmare. As a hold-up horse (IPS 4/5) heading into a crawling pace on the Newcastle tapeta, she will need a monumental turn of foot to overcome the mathematical disadvantage of her track position.
3. Dreamasar (97 Points) — The "Class Move" Sniper
While her total score is a distant third, she is the only horse near the top of the market to trigger Trend #1: Class Move (40% SR, Advantage 35) and Trend #2: OR v High (36.36% SR, Advantage 31). She is stepping up in grade exactly the way past winners have, validating her as the sharpest each-way alternative.
Updated Tactical Verdict
The integration of the Trend Hunter data solidifies the favourite while creating a clear roadmap for structuring exotic bets.
The Cast-Iron Anchor: Diamond Rain (1.67). It is rare to see a horse tick every single box: raw class, pace-map immunity, and overwhelming historical trend alignment. She is a completely justified odds-on favourite.
The Exacta Value: Dreamasar (12.0). She hits the two most heavily weighted individual trends in the race (Class Move and OR v High) and sits in the tactical sweet spot (IPS 2/3) to avoid the pace trap that threatens the closers.
The Risk/Reward Lay: Sky Safari (5.50). Laying a horse with a 250-point trend score carries risk, but laying a hold-up horse in a "Very Weak" pace scenario is a strictly mathematical play. If the front-runners crawl, she will simply be too far back to catch Diamond Rain.
The Tactical Wildcard: Ashariba (26.0). Her trend score (77) is poor, but she remains the sole front-runner (IPS 1). If you are building trifectas, she is the mathematical pace anomaly to throw in for third.

 

[Paul Brown]

ai. I was interested it in what happened to chats if I deleted them and what was tempory mode.
The answer was as follows ;
What are tempory chats ?




Temporary chats are a privacy-focused feature designed for one-off tasks or sensitive queries where you don't want to leave a digital footprint.
When you use a temporary chat, the system handles your data differently to ensure a contained, private environment:

• No Chat History: The conversation is not saved to your history sidebar, so it won't appear in your list of past chats.
• No Future Influence: The details discussed in a temporary chat are completely isolated—they will not be used to train future AI models, nor will they influence your future interactions or personalized context.
• Disabled Integrations: To keep the environment fully secure and private, external extensions and custom Gems are automatically disabled while this mode is active.

It is essentially a "private browsing" mode for your AI assistant—ideal for when you need to brainstorm a sensitive topic, draft a quick email, or ask a one-off question without cluttering your chat history.





where do I turn on Tempory chat and how do i Go back to a chat recorded in my history mode




Whether you are using Google Gemini or ChatGPT, starting a temporary chat and switching back to your standard history mode is incredibly easy.
Here is exactly how to do both on each platform.

Google Gemini​

1. How to turn ON Temporary Chat​

• On Web: Look at the top-left corner of the page next to "New chat" (or in the top-right corner near your profile). Click the dashed chat bubble icon (or the pencil/temporary chat icon) to enter the private mode.
  Information and Technology Services - University of Michigan+ 1
  
  
• On Mobile: Tap the Menu icon (three horizontal lines) in the top-left corner, then tap Temporary chat from the sidebar.
  YouTube
  

How to know it's on: The header at the top of your screen will clearly display "Temporary chat".
YouTube

2. How to go back to History Mode​

To leave the temporary chat and return to your saved history:

• Click the "X" icon on the temporary chat indicator at the top of the screen, or simply click New chat in the left sidebar.
  Information and Technology Services - University of Michigan
  
  
• Any saved chats from your past history will immediately reappear in the "Recents" panel on the left sidebar.
  Information and Technology Services - University of Michigan
  

ChatGPT​

1. How to turn ON Temporary Chat​

• On Web & Mobile: Start a new chat, then click on the model selector at the top-left of your screen. In the dropdown, toggle the "Temporary Chat" switch to ON (or click the "Temporary" pill button in the top-right corner).
  OpenAI Help Center
  

How to know it's on: The entire chat interface will have a translucent, stylized frame with a prominent "Temporary Chat" label at the top.

2. How to go back to History Mode​

• To exit, click the "New chat" button in the left sidebar or toggle the "Temporary Chat" switch back to OFF in the model menu.
  Guideflow
  
  
• Once clicked, your previous, recorded chat history will instantly populate the left sidebar again, allowing you to click on and resume any of your old conversations.

Important Reminder: Once you close a temporary chat tab or switch away from it, the conversation is gone forever. You cannot convert a temporary chat back into a saved chat later, so make sure to copy any important code, drafts, or information before you exit!
Information and Technology Services - University of Michigan+ 1

If i delete a chat can it still be use by developers to train ai




The short answer is yes, potentially, but it depends entirely on your settings before you deleted the chat.
Simply clicking the "Delete" trash can icon on a chat does not automatically stop the AI company from using it for training. Here is how it actually works behind the scenes for both OpenAI (ChatGPT) and Google (Gemini).

1. ChatGPT (OpenAI)​

If you have ChatGPT's default settings turned on, OpenAI may use your conversations to train future models.
Anarlog

• If you delete a chat without changing your settings: OpenAI will keep the deleted conversation on their servers for up to 30 days before permanently purging it. During or before that 30-day window, if the data has already been "de-identified" (stripped of personal details) and queued for training, it can still be used. Deleting the chat after the fact does not "recall" it from the training pool.
  OpenAI+ 1
  
  
• How to stop it: Go to Settings > Data Controls and toggle off "Improve the model for everyone". Once this is off, future chats will not be used for training, even before you delete them. (Note: This does not retroactively remove chats that have already been trained on.)
  Anarlog
  

2. Gemini (Google)​

Google has a slightly different approach, linking your chat history and your training permissions together.
Reddit

• If you delete a chat without changing your settings: If your Gemini Activity is turned on, Google uses those chats for training. Furthermore, a small portion of chats are selected for human review to improve the AI. If your chat was reviewed by a human, deleting it will not remove it; Google retains human-reviewed chats (disassociated from your account) for up to 3 years.
  PCMag UK+ 2
  
  
• How to stop it: Go to your Gemini Apps Activity settings and turn the activity toggle OFF. Doing this stops Google from using your chats for training, but it also means Gemini will no longer save your chat history in the sidebar.
  PCMag UK+ 1
  

The Golden Rule for Privacy​

If you want to ensure a conversation is never used for training, you must take action before you send the first message:

1. Turn off training/activity settings in your account options.
   PCMag UK
   
   
2. Use Temporary Chat mode, which automatically bypasses the training pipeline entirely and deletes the data from the servers within 30 days.
   OpenAI
   
   
3. Never type highly sensitive information (like passwords, credit card numbers, or proprietary work data) into any cloud-based AI, regardless of your settings.

So if you had a killer system for horseracing some nosy person in some faraway place might land on it.
As i use a new chat for each horserace, my sidebar can soon fill up. But its there in the background for 3 years. I do need to tell you that I don'thave the magic formula though !