Introducing Ford AI Assistant

[bc1]

I was making one of my usual jokes earlier but I generally use google AI to ask questions including those about my truck which I'm still learning. If Doug is available now then I could sure use him, just tell me where, even on a beta/test mode. Maybe he can help with my other ford f150 ice pickup.

Thanks Brian. I know Ford is trying to do what they can in light of everything going on in the world and their shareholders and customers. Also I am very happy with my new 2025 flash just bot on Sep 30 on the last day of the tax credits from Blue Springs Ford in Kansas City who treated me very well and honestly in the process including their sales staff and sales manager so cudos to them.

Sponsored

 

[MaintGrl]

If you are an expert in anything, you know that publicly-available ai isn’t an expert in anything and cannot be an expert in anything.

Just 0.001% bad data (one bad or malicious data token in every 100,000 data tokens) in AIs means you have a system that pushes misinformation and disinformation and creates more harm than good.

Medical ais ingest millions to billions of pieces of data from nobody knows what sources (because though they’re happy to steal form anyone and everyone, they won’t share who they’ve stolen from), including the cesspool of the internet at large.

I regularly reject all ai in my medical interactions.


https://www.nature.com/articles/s41591-024-03445-1

 

[SomeStupidDriver]

I’m not sure why you’re white knightning for Brian……. Don’t get me wrong, I’m grateful for his insight and presence in the forum, but I’m sure he’s quite capable of answering on his own without your intervention.

A few additional things:
1. Embargo’s are for 3rd parties, not employees of Ford.
2. Brian has made comments on some of these issues over the last few months.
3. Personal information data concerns? Nobody asked him to share client data. This is a non-issue.
4. Legal concerns are addressed as easily as Brian saying he can’t comment at this time.

The questions asked are pretty reasonable and some of which impact EVERY Lightning owner.

People ask the same questions multiple times a day in all kinds of threads. It would absolutely shit up every single thread if the information hasn't changed, and Ford was just spamming "no comment" or "nothing new" all day, every day, to every single person that didnt understand that there usually isn't anything new to report since the last time they answered the question.

It's not like he's got all this new information and answers to the questions we've been asking and he's just decided to keep it to himself. It's not white knighting to ask people to have some common sense.

 

[ClevelandBeemer]

People ask the same questions multiple times a day in all kinds of threads. It would absolutely shit up every single thread if the information hasn't changed, and Ford was just spamming "no comment" or "nothing new" all day, every day, to every single person that didnt understand that there usually isn't anything new to report since the last time they answered the question.

It's not like he's got all this new information and answers to the questions we've been asking and he's just decided to keep it to himself. It's not white knighting to ask people to have some common sense.

It is very much white knightning. Brian can and did speak for himself. Brian is great and overall his communication has also been fantastic. I’m not being critical of Brian, I’m calling for action from Ford on outstanding issues.

You call it having common sense, I call it making assumptions. I don’t know what Brian and Ford are thinking, so why would I assume an answer? I follow the forum pretty closely and feel confident I’d catch if Brian made a statement on the questions asked.

Furthermore, Ford is choosing to spend R&D on a feature that’s not even in cars yet while issues that have persisted for YEARS on the Lightning have gone unaddressed. If you don’t see the problem there then I don’t know what to tell you.

 

[Rocinante]

If I cannot shut it off entirely I am likely to switch brands. I do NOT want AI sending Ford, or anyone else, any information about me as it is.

 

[chriserx]

If I cannot shut it off entirely I am likely to switch brands. I do NOT want AI sending Ford, or anyone else, any information about me as it is.

This is already happening, without a filter unfortunately. However, there are currently toggles to 'disable' this dissemination. Whether or not they actually do this, it's hard to really say. One of the many reasons I like to have all of my data routed through my personal VPN. Just wish I could do it with the onboard modem too.

 

[Rocinante]

This is already happening, without a filter unfortunately. However, there are currently toggles to 'disable' this dissemination. Whether or not they actually do this, it's hard to really say. One of the many reasons I like to have all of my data routed through my personal VPN. Just wish I could do it with the onboard modem too.

I hear you. One more thing listening to us constantly and sharing info without permission (since they make you give a blanket permission just to use the app itself).

 

[Temix]

IMHO, this is a textbook example of big-corporation thinking: spend another billion to become an “AI company” and end up with yet another product nobody actually uses.

 

[22legit2quit]

Too many irons in the fire here Fomoco. Can you fix or improve blue cruise? No? Alright another useless item that 99% of people won’t use outside of the first time and say “this is cool, but…”.

This is a monetized data grab in an effort to make a buck so no thanks.

 

[Jim Lewis]

If you are an expert in anything, you know that publicly-available ai isn’t an expert in anything and cannot be an expert in anything.

Just 0.001% bad data (one bad or malicious data token in every 100,000 data tokens) in AIs means you have a system that pushes misinformation and disinformation and creates more harm than good.

1. Types of Data Used in Medical AI

Curated Clinical and Scientific Data
In clinical and regulatory-grade medical AI systems (especially those used in diagnostics, medical imaging, decision support, etc.):

• Datasets are carefully curated from trusted sources: peer-reviewed journals, clinical trial data, annotated patient records, expert-reviewed cases, hospital electronic health records (EHRs), medical imaging from controlled sources, and other verified biomedical information.
• These datasets are usually cleaned, annotated by domain experts, cross-checked, and often harmonized with medical coding standards (ICD, SNOMED, etc.).
• This curation improves veracity and clinical relevance of the model’s knowledge.
• These models are trained in a supervised or semi-supervised way with clinical outcomes and labels, not just raw Internet text.

Many formal medical AI tools in use today (e.g., FDA-cleared diagnostic AI for imaging) are trained this way and continuously evaluated in clinical environments.
2. What the Nature Medicine Paper Shows
The Nature Medicine study you linked (Alber et al., 2025) specifically looked at large language models (LLMs) that are pre-trained on open-internet data like The Pile — which aggregates huge amounts of web text (including Common Crawl) without medical curation.
Key findings of that paper

• Web-scale training data contain a mix of medical text, some from PubMed but a significant fraction (e.g., ~27% of medical terms) comes from sources that are not medically curated and therefore vulnerable to misinformation.
• A tiny amount of poisoned or false content injected into such training sets can measurably increase the model’s tendency to produce harmful medical outputs, even though performance on common benchmarks looks the same.
• The risk exists not only from deliberate poisoning but also from the “incidental” presence of outdated, pseudoscientific, or poorly-sourced medical information on the open web.

This paper essentially highlights a vulnerability in large, general-purpose LLMs trained on unconstrained web text — not the curated medical-AI systems used in regulated clinical decision support.
3. Real Medical AI Systems vs. General Large-Language Models
General LLMs (e.g., GPT, Gemini, Claude trained on web data)

• Trained on broad internet data that includes professional content but also blogs, forums, non-expert text, and potentially false/misleading medical claims.
• They can hallucinate or extrapolate incorrectly, especially when prompted outside their “core knowledge” or without grounding mechanisms.
• These are not appropriate for autonomous medical decision-making without safeguards.

These general models can be adapted for medical use, but risk remains if they aren’t specifically trained and validated on curated clinical datasets.
4. How Safety and Accuracy Are Improved in Medical AI
For medical applications to be safe and clinically useful, developers use one or more of the following strategies:
Curated Training Data

• Using datasets from clinical practice, annotated by healthcare professionals.
• Restricting sources to verified scientific and clinical repositories.

Fine-tuning and Supervision

• Models can be fine-tuned on medically accurate corpora with expert labeling, which greatly reduces hallucinations.

Knowledge Augmentation / Retrieval

• Retrieval-augmented systems link generative output to trusted medical knowledge bases (e.g., PubMed, UpToDate, structured clinical knowledge graphs), enhancing verifiability.

Safety Layers & Verification

• Models may be wrapped with validators — e.g., knowledge graphs or rule-based checks — to screen outputs against known clinical facts (described as a mitigation strategy in the Nature Medicine paper).

Human-in-the-Loop

• Final medical decisions involve clinicians — AI assists but does not replace professional judgment.

5. Bottom Line

System Type	Data Source	Veracity Risk
Clinical Medical AI (regulated, domain-specific)	Curated clinical/peer-reviewed datasets	Low (with ongoing checks)
General Internet-trained LLMs	Web-scale, not curated	Higher risk of misinformation
LLMs fine-tuned for medical use	Combination of curated and web data	Medium–low depending on process

So:

• **Medical AI **aims to use carefully vetted data where possible.
• General AI models trained on internetwide text do carry the risk highlighted in that Nature paper — especially absent strong curation or safeguards.

Hence, in practice, high-stakes clinical tools rely on curated datasets and validation rather than unconstrained internet data, but pure LLMs without such controls can and do suffer from data integrity issues.

The next time you grab something off the Internet, you might want to examine it in more in depth...

Full-analysis: https://chatgpt.com/share/696164b3-5234-800a-a513-183946965a53

If anyone just relies on ChatGPT, Google Gemini, MS Copilot, or Anthropic, etc., for medical advice, they're not being very smart. Medical AI is NOT ChatGPT.

Also, anyone opposed to AI in medicine might want to check what their PCP is doing these days. Many large clinical practices in San Antonio are now using AI to record patient-PCP conversations and produce an appointment summary (to be reviewed and edited by the PCP) for your medical records. It's not necessarily that the AI summary is so superfantastic (although a physician neighbor of mine who uses it in his practice extols it), but it's a cost/benefit thing in the medical budget squeeze.

 

[ClevelandBeemer]

The next time you grab something off the Internet, you might want to examine it in more in depth...

Full-analysis: https://chatgpt.com/share/696164b3-5234-800a-a513-183946965a53

If anyone just relies on ChatGPT, Google Gemini, MS Copilot, or Anthropic, etc., for medical advice, they're not being very smart. Medical AI is NOT ChatGPT.

Also, anyone opposed to AI in medicine might want to check what their PCP is doing these days. Many large clinical practices in San Antonio are now using AI to record patient-PCP conversations and produce an appointment summary (to be reviewed and edited by the PCP) for your medical records. It's not necessarily that the AI summary is so superfantastic (although a physician neighbor of mine who uses it in his practice extols it), but it's a cost/benefit thing in the medical budget squeeze.

So your big revelation is AI dictation and meeting summary? Copilot has done this for some time. It’s also not infallible and misses items, but overall gets you 90% there.

Think bigger. For example there are AI solutions being tested to do device clinic remote monitoring work that can detect rhythms and grade on low or high risk in order to prioritize what humans review for follow-ups and programing changes. Powerful, time saving, and should prioritize those who need immediate follow-up.

HOWEVER for your vehicle, this is a glorified search engine and should be far from a priority of where to direct R&D resources.

 

[Jim Lewis]

Think bigger. For example there are AI solutions being tested to do device clinic remote monitoring work that can detect rhythms and grade on low or high risk in order to prioritize what humans review for follow-ups and programing changes. Powerful, time saving, and should prioritize those who need immediate follow-up.

HOWEVER for your vehicle, this is a glorified search engine and should be far from a priority of where to direct R&D resources.

I think in the quoted ChatGPT reply I provided, it's pretty evident that medical AI is a vast encompassing topic that can involve in-depth analysis, or what's radiographic image analysis all about!? I also suffer from atrial fibrillation (a common failing in old age) and have a Kardia Mobile device, and know that the ECGs generated by that can be interpreted by AI. I'm just too cheap to subscribe to their expensive AI service. (I also have a Medtronics loop recorder embedded in my chest whose recordings are also computer-analyzed, but I don't know if AI is involved with that).

I think forum members might be misinterpreting why Ford is offering an AI Assistant in its vehicles. If it means fewer people minding phones at call centers to answer customers' questions, the cost-savings behind that (fewer salaries, medical and retirement plans to pay out) may be all that's involved, not the desire to wow us with AI. Just like AI summarizing PCP office exam results - follow the MONEY...

 

[ClevelandBeemer]

I think in the quoted ChatGPT reply I provided, it's pretty evident that medical AI is a vast encompassing topic that can involve in-depth analysis, or what's radiographic image analysis all about!? I also suffer from atrial fibrillation (a common failing in old age) and have a Kardia Mobile device, and know that the ECGs generated by that can be interpreted by AI. I'm just too cheap to subscribe to their expensive AI service. (I also have a Medtronics loop recorder embedded in my chest whose recordings are also computer-analyzed, but I don't know if AI is involved with that).

I think forum members might be misinterpreting why Ford is offering an AI Assistant in its vehicles. If it means fewer people minding phones at call centers to answer customers' questions, the cost-savings behind that (fewer salaries, medical and retirement plans to pay out) may be all that's involved, not the desire to wow us with AI. Just like AI summarizing PCP office exam results - follow the MONEY...

Yep I’m familiar with your hardware, or more specifically my wife is, as it’s her profession. You Medtronic’s unit is exactly what I’m referring to on the AI pre-screens. Depending on the your provider, expect to see this coming soon to their monitoring process. I’m not sure you’d ever know that they’re utilizing AI screening for their remotes unless you ask them.

I’m in total agreement, it’s always about the money. However I don’t think vehicle owners are going to use this service like they think. I also don’t think it’s going to reduce/remove humans from customer service. It’s just the new hot thing to add AI assistants to vehicle or app interfaces. BMW’s new assistant on the IX3 is a great example of this.

 

[Maxx]

How about giving me back the charging info/functionality that is taken hostage to get my driving info instead?

 

[TechnoSwiss]

I'm still waiting for it to recognize when I ask it to turn the rear defrost on with the hands free because that button is hard to hit while driving and keeping an eye on the road.

Sponsored