Subject: Re: MS Edge was never useful but at least it had worked for Copilot but what AI/LLM now?

On 8/26/26 2:02 PM, Maria Sophia wrote:

MS Edge has never been useful so I used it solely for Copilot AI/LLM.
But that died a few weeks ago in terms of no-registration searches.

I've since then updated my MS Edge taskbar-pinned target to open these.

"C:\Program Files (x86)\Microsoft\Edge\Application\msedge.exe" --profile-directory=Default edge://settings/privacy/clearBrowsingData https://gemini.google.com/app https://chatgpt.com/ https://www.perplexity.ai/ https://duck.ai/ https://plai.chat/

Where the use model is if I need to look up something, using natural
language terms, I pop open the MS Edge icon pinned to the taskbar.

And then I ask something like:
"Why do people buy cellphones that have no sd card slots?"
"Why do people put premium gasoline into a Honda Civic?"
"Is the boron in laundry product the same as in the ant bait?"
etc.

When one AI/LLM times out, I move to the next tab, and so on.
Sometimes I even ask all the tabs the same question, if it's complex.
"Why do people think pH has any major effect on pool chemistry?"
"What is the closest analogy to the illegal war in Ukraine?"
"What does Windows 10 not do that Windows 11 does that's useful?"
etc.

Then I can have the answers from one AI/LLM rebut those of the other.
But I only do that for deep serious technical queries, as it's costly.

My question for the group is what other free no-registration AI/LLMs
can we add to the otherwise useless MS Edge web browser invocation?

I use copilot but not enough to be concerned about. I do some days get into a long Q&A
with it and get kicked out 'you've hit your limit for one day'. Good thing is I can pick
up tomorrow, but this is like 1 in 8 sessions.

--
Mint 22.3, Thunderbird 153.1.0esr, Firefox 154.0
Alan K.

--- PyGate Linux v1.5.19
* Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)

Subject: Re: MS Edge was never useful but at least it had worked for Copilot but what AI/LLM now?

On Wed, 8/26/2026 4:27 PM, Alan K. wrote:

On 8/26/26 2:02 PM, Maria Sophia wrote:

MS Edge has never been useful so I used it solely for Copilot AI/LLM.
But that died a few weeks ago in terms of no-registration searches.

I've since then updated my MS Edge taskbar-pinned target to open these.

"C:\Program Files (x86)\Microsoft\Edge\Application\msedge.exe" --profile-directory=Default edge://settings/privacy/clearBrowsingData https://gemini.google.com/app https://chatgpt.com/ https://www.perplexity.ai/ https://duck.ai/ https://plai.chat/

Where the use model is if I need to look up something, using natural
language terms, I pop open the MS Edge icon pinned to the taskbar.

And then I ask something like:
ÿ "Why do people buy cellphones that have no sd card slots?"
ÿ "Why do people put premium gasoline into a Honda Civic?"
ÿ "Is the boron in laundry product the same as in the ant bait?"
ÿ etc.

When one AI/LLM times out, I move to the next tab, and so on.
Sometimes I even ask all the tabs the same question, if it's complex.
ÿ "Why do people think pH has any major effect on pool chemistry?"
ÿ "What is the closest analogy to the illegal war in Ukraine?"
ÿ "What does Windows 10 not do that Windows 11 does that's useful?"
ÿ etc.

Then I can have the answers from one AI/LLM rebut those of the other.
But I only do that for deep serious technical queries, as it's costly.

My question for the group is what other free no-registration AI/LLMs
can we add to the otherwise useless MS Edge web browser invocation?

I use copilot but not enough to be concerned about.ÿÿ
I do some days get into a long Q&A with it and get kicked out> 'you've hit your limit for one day'.ÿÿ Good thing is I can
pick up tomorrow, but this is like 1 in 8 sessions.

I knew it had an attitude, but I didn't know it had a limit :-)

It always half-answers your question, in an effort to keep the
chat going.

You can run them locally, but you have to be patient. For one request
to write a program, and with high reasoning enabled, it took 50 minutes
to half-write the program. The high reasoning took about eight minutes
of that, and the vast majority of the time was spent on token output.

They tell me it would use a video card for token output, *if* the question
fits entirely within the video card. But with my puny hardware resources, it always uses the CPU for token output, which can be slow.

For about $15K, I could upgrade enough to be within reach of a sorta-good result (500 watts).
For $150K, I could "hit it out of the park" (GB300 tower 1600 watts). For maybe $8 million
(Cerebras single node 400K cores 20kW electricity max), it could answer questions
faster than you could ask them (3000 tokens per second best case). That price estimate is
based on a 64 node max-sized install costing $500 million. You can see
how they've "covered every reasonable price point". In gold mining days,
these would be the solid gold mules and the solid gold shovels.

And my dreams of AMD designing a "good" accelerator are smashed. It seems HBM3e is
about 5x more expensive than DDR5 memory or GDDRx, so attempting to go toe-to-toe
with the AI industry isn't going to work. The accelerator card just becomes more expensive, while the "memory" is more available (at the high price).

GDDRx HBM3e (more available, more expensive)
| ==> |
GPU GPU

Not all the accelerators, do the same math. If you have the luck to get
an accelerator that has a single format like INT8, then your model has
to be quantized (maybe, even on the fly), before the accelerator can work on it.
Some devices might support five formats, others only one (really cheap shit). And in true FUD fashion, one product was claimed to be unable to be
sharded (problem split across hardware cards) -- but maybe a competitor
was making the claim. You can't trust people on the Internetz for such
nuggets. As it is, no two people seem to get the same token rate, out
of the same hardware, suggesting "too many knobs" syndrome.

Paul


--- PyGate Linux v1.5.19
* Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)

Subject: Re: MS Edge was never useful but at least it had worked for Copilot but what AI/LLM now?

On 2026-08-27 03:20, Paul wrote:

You can run them locally, but you have to be patient. For one request
to write a program, and with high reasoning enabled, it took 50 minutes
to half-write the program. The high reasoning took about eight minutes
of that, and the vast majority of the time was spent on token output.

They tell me it would use a video card for token output,*if* the question fits entirely within the video card. But with my puny hardware resources, it always uses the CPU for token output, which can be slow.

There is a chip that can accelerate the ai more than gpus. I read about
it in the ieee spectrum magazine, but I don't remember the name of the
thing. Laptops will come with that hardware so they can handle ai
locally with reasonable battery usage.

probably asking an ai about this it will tell better details than me :-D

--
Cheers, Carlos.
ES??, EU??;

--- PyGate Linux v1.5.19
* Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)

Subject: Re: MS Edge was never useful but at least it had worked for Copilot but what AI/LLM now?

On Thu, 8/27/2026 3:03 AM, Carlos E.R. wrote:

On 2026-08-27 03:20, Paul wrote:

You can run them locally, but you have to be patient. For one request
to write a program, and with high reasoning enabled, it took 50 minutes
to half-write the program. The high reasoning took about eight minutes
of that, and the vast majority of the time was spent on token output.

They tell me it would use a video card for token output,*if* the question
fits entirely within the video card. But with my puny hardware resources, it >> always uses the CPU for token output, which can be slow.

There is a chip that can accelerate the ai more than gpus. I read about it in the ieee spectrum magazine, but I don't remember the name of the thing. Laptops will come with that hardware so they can handle ai locally with reasonable battery usage.

probably asking an ai about this it will tell better details than me :-D

Acceleration devices, have support for just one numeric format,
or for multiple numeric formats.

Model files, some of them come pre-quantized to a particular numeric
format, for a reason. If your hardware happens not to have that format,
then something must be done to fix that. This is why there are two
overlapping formats, one an open thing, the other an NVidia version.
If a model comes in the NVidia version, there's a message there for you.

Acceleration devices tend to quote the TOPS figure for their "fastest" format. Whether they even have a second format, is another question.
Maybe an Intel NPU, they quote the INT8 performance. Now, in LMStudio,
look in the model list, and how many of the models are INT8 ?
Quantizing a model, can break it, so the results are not necessarily
good.

That's why, the single TRIT model that was made, it was "trained in TRITs"
and "Inference is in TRITs". That is to ensure that quantization does
not ruin it. A CPU can handle [-1,0,+1 ] as a representation, but
typical acceleration devices do not have a hardware block for that.

This makes the topic of models and accelerators, a bit of a mine field.
There are things we could be doing... which are not happening. A TRIT
model requires a lot less RAM. And, there are none.

You can see a discussion of what happens when dishing out resources on big machines.

https://aimultiple.com/llm-quantization

"Int4 degrades code generation more than knowledge"

This is why I use code generation, as my "acceptance test" for local models. Can it write code ? Is it timid, like one model which only wrote comment
text for the source file ? I don't particularly care, if it can
list all the flavors of toothpaste. That should be pretty easy for it.
Even a small small model, could tell you about peppermint toothpaste.

Paul

--- PyGate Linux v1.5.19
* Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)

Subject: Re: MS Edge was never useful but at least it had worked for Copilot but what AI/LLM now?

On 08/27/2026 3:03 AM, Carlos E.R. wrote:

On 2026-08-27 03:20, Paul wrote:

You can run them locally, but you have to be patient. For one request
to write a program, and with high reasoning enabled, it took 50 minutes
to half-write the program. The high reasoning took about eight minutes
of that, and the vast majority of the time was spent on token output.

They tell me it would use a video card for token output,*if* the question
fits entirely within the video card. But with my puny hardware
resources, it
always uses the CPU for token output, which can be slow.

There is a chip that can accelerate the ai more than gpus. I read about
it in the ieee spectrum magazine, but I don't remember the name of the thing. Laptops will come with that hardware so they can handle ai
locally with reasonable battery usage.

probably asking an ai about this it will tell better details than me :-D

Iirc,
NPU - neural processing unit
ASIC - app specific
- i.e. for a specific AI task

Also another older type of chip/circuitry based on gate array(?) and re-programmable.

--
...w­¤?ñ?¤

--- PyGate Linux v1.5.19
* Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)

Subject: Re: MS Edge was never useful but at least it had worked for Copilot but what AI/LLM now?

On Thu, 8/27/2026 12:38 PM, ....winston wrote:

On 08/27/2026 3:03 AM, Carlos E.R. wrote:

On 2026-08-27 03:20, Paul wrote:

You can run them locally, but you have to be patient. For one request
to write a program, and with high reasoning enabled, it took 50 minutes
to half-write the program. The high reasoning took about eight minutes
of that, and the vast majority of the time was spent on token output.

They tell me it would use a video card for token output,*if* the question >>> fits entirely within the video card. But with my puny hardware resources, it
always uses the CPU for token output, which can be slow.

There is a chip that can accelerate the ai more than gpus. I read about it in the ieee spectrum magazine, but I don't remember the name of the thing. Laptops will come with that hardware so they can handle ai locally with reasonable battery usage.

probably asking an ai about this it will tell better details than me :-D

Iirc,
NPU - neural processing unit
ASIC - app specific
ÿ- i.e. for a specific AI task

Also another older type of chip/circuitry based on gate array(?) and re-programmable.

There is a diagram of an NPU here.

https://intel.github.io/intel-npu-acceleration-library/npu.html

An NPU can be 50 TOPS. a GPU can be 1000 TOPS, but the operations per second depend on the representation (INT8, FP16). The FP16 rate could be much lower. An NPU might not even have FP16 capability. The CPU can do any math you can think of, but at quite-low TOPS rates.

Software loading models, does not "dispatch" onto all your disparate acceleration devices. The CPU manufacturers like to "total" the TOPS numbers from the three accelerators they've got. Whereas the model running software laughs at this silliness, and only uses the most-productive acceleration
device and it ignores the others. If you include a slow device and shard
a calculation, the slow device holds up everybody. Dispatching is not
nearly as optimistic as Intel and AMD would have you believe.

This is one of the reasons, that NPU graph in Task Manager on your new laptop, it might never ever lift off the peg. For the NPU to "light up", someone
has to go out of their way, to use it. They have to think of a specific application, make a model of the same type as the single format the
NPU supports, hope there is a driver for the NPU, *then* load it up and
run it. That means a company like Intel, writes that program :-)
The NPU then, is a kind of bar bet. "I betcha I can make that line move...
Hold my beer"

Paul


--- PyGate Linux v1.5.19
* Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)

Subject: Re: MS Edge was never useful but at least it had worked for Copilot but what AI/LLM now?

On 2026-08-27 18:38, ....winston wrote:

On 08/27/2026 3:03 AM, Carlos E.R. wrote:

On 2026-08-27 03:20, Paul wrote:

You can run them locally, but you have to be patient. For one request
to write a program, and with high reasoning enabled, it took 50 minutes
to half-write the program. The high reasoning took about eight minutes
of that, and the vast majority of the time was spent on token output.

They tell me it would use a video card for token output,*if* the
question
fits entirely within the video card. But with my puny hardware
resources, it
always uses the CPU for token output, which can be slow.

There is a chip that can accelerate the ai more than gpus. I read
about it in the ieee spectrum magazine, but I don't remember the name
of the thing. Laptops will come with that hardware so they can handle
ai locally with reasonable battery usage.

probably asking an ai about this it will tell better details than me :-D

Iirc,
NPU - neural processing unit

That's the one I was thinking about.

ASIC - app specific
ÿ- i.e. for a specific AI task

Also another older type of chip/circuitry based on gate array(?) and re- programmable.

--
Cheers,
Carlos E.R.
ES??, EU??.

--- PyGate Linux v1.5.19
* Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)

Subject: Re: MS Edge was never useful but at least it had worked for Copilot but what AI/LLM now?

On Sat, 8/29/2026 3:49 PM, Andy Burns wrote:

....winston wrote:

Carlos E.R. wrote:

There is a chip that can accelerate the ai more than gpus. I read about it in the ieee spectrum magazine, but I don't remember the name of the thing. Laptops will come with that hardware so they can handle ai locally with reasonable battery usage.

probably asking an ai about this it will tell better details than me :-D

Iirc,
NPU - neural processing unit

I encountered a Lenovo laptop with ARM Snapdragon CPU and NPU last week (neighbour bought it for his daughter).

Remembering that Win11 was supposed to enable "Recall" if it detected an NPU, I did check, but it seems Microsoft reversed their decision to enable it by default if an NPU was present ...

For Recall, you have to Opt-In to get it.

The last review I saw, claimed that privacy issues
were still not handled with 100% accuracy.

Paul





--- PyGate Linux v1.5.19
* Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)

Subject: Re: MS Edge was never useful but at least it had worked for Copilot but what AI/LLM now?

On Sat, 8/29/2026 5:43 PM, Andy Burns wrote:

Paul wrote:

Andy Burns wrote:

Remembering that Win11 was supposed to enable "Recall" if it detected an NPU, I did check, but it seems Microsoft reversed their decision to enable it by default if an NPU was present ...

For Recall, you have to Opt-In to get it.

You certainly do now, but as I recall (!), the original version would be opt-out?

Someone at Microsoft also thought the original version worked.

The reviewer feedback said otherwise. The product is supposed
to blot out credit card numbers. Or maybe a password field,
if the password field is not the obscured-entry type.

We don't know what happens, if a person visits a web page,
the web page has a

Bobby Drop Tables

text string, and the AI executed it. The technology might have
an attack surface that is hard to protect.

It's one thing to have a running AI, it is something else to be
feeding it unpredictable inputs.

Paul

--- PyGate Linux v1.5.19
* Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)