howard.mov
Region: US
Sunday 19 April 2026 00:02:21 GMT
109301
7150
81
703
Music
Download
Comments
Vision :
Bro we’ve BEEEN doing this - check out Hermes Agent
2026-04-22 03:22:06
18
Ryan :
Next will be the meta meta harness.
2026-04-19 23:00:17
11
void star :
Everyone is working on this
2026-04-19 03:05:35
109
nicholas chin :
This is another AI Ranting. But it involves the control layer for calling tools agents now this they have gone to even this kind of level where they are using tools which is a set of instruction that is specialised to do certain specific cars or certain hard coded algorithm that could be used efficiently by the agent which is a robot that is in digital information for to do something so you have you need to have a control layer on that and up to recently even now the control layer is hard coded by human algorithm with no regressive feedback nature to optimise how efficient it can be. So the train and other neural network to optimise that control layer to somehow like for example if you want to read a PDF and annotate with information you might want to write a code for it or you might just want to call the native to in available PDF open source format to just simply apply so this is the kind of optimisation that this approach is doing. And is it innovative I will say no. It is an error that was under optimised and the reason why it is under optimised is because it is as the term ecstoric wish I think I spelt wrongly but I cannot remember how to spell but it means that it is an error that you know people sell them go into. For this case people sell them play with the architecture you can build model you can trade model but you don't actually design the architecture then you don't actually then do the structure which is the hard coding so unless you are a neural network architecture designer which is by itself rather rare, you won't be able to be thinking about optimising this area. But as if you are one such you have a bird's eye view most probably you'll be thinking of optimising that one sooner or later. That is the reason why I think this is not innovative but it is an optimisation of a structure of the model that arises due to the architectural requirement. It is not a wrong use of term I think this terms are quite accurate structure arising from architecture requirement.
2026-04-19 14:03:05
2
iCreatr :
Meta is crap…they’re so behind
2026-04-20 08:20:53
0
X :
What is a harness ? AI Wrapper ?
2026-04-20 04:28:54
0
Connor Tierney :
This should just be how the models work to begin with. Now that compute and energy are maxing out, they will need to focus on efficiency updates.
2026-04-24 18:28:00
1
Kaeleon :
I just got done building my obsidian vault with hermes agent :)
2026-04-24 00:52:54
1
DQ Blizzard of Oz 🎰 :
How does this compare to hermes?
2026-04-20 00:35:48
0
Viral Internet Highlights :
I'm so sorry can to explain this like I'm 5 cause I don't really understand
2026-04-21 19:01:12
1
Lucid Reverie :
more than 6x
2026-04-19 07:45:27
0
Hendrik :
Correct.. token drift solved up to 60%
2026-04-19 16:33:13
7
CRYINGAI :
Bubble so big we making up frameworks now 😂
2026-05-03 19:18:01
1
gool, gabagool :
it’s too expensive to automate
2026-04-24 21:01:38
0
AB :
I like droid as a harness
2026-04-19 07:08:17
0
1cj937 :
Is this not what langsmith does with langchain and langgraph
2026-04-23 20:51:20
2
Bill Kellaway :
I'm following you I'm very interested
2026-04-19 04:06:00
0
The AI Dividend :
This is the current frontier.
2026-04-19 14:07:35
2
Auge Eidos :
Openspec.
2026-06-08 11:08:37
0
MojoScott :
I focus on the orchestrator harness and don’t even let the model harness see this shit.
2026-06-05 02:19:45
0
Daniel O :
So do you need to run this with another harness or is this a self-contained harness?
2026-06-04 11:44:59
0
AI For People Tired of Noise :
Are there any repos on this? 🔥
2026-06-03 04:20:10
0
bizkut0 :
how long until we get a meta meta harness?
2026-06-03 01:47:20
0
Megan Writes :
Yesss this is great!
2026-05-31 14:25:16
0
David :
Interested
2026-04-19 08:44:42
0
To see more videos from user @howard.mov7, please go to the Tikwm
homepage.
