GPT-5.6 is the first model where I’ve actually frustrated to use it.
I’m explicitly telling it to do something extremely specific and it’s just not listening to me.
Eg, I gave it an image to update. The image is sized 400x200 pixels. It then generates a new image at 300x300. I explicitly state to be 400x200 in size and it won’t listen.
So, you have these models that you put things into (like text), and get things out of.
For many models, you can include a "tool definition" in that text input. Remember this definition is just JSON :)
Let's say you send a tool definition to the model, plus a text question.
The model responds with a request for you to call the tool. It's also JSON.
What shall you, the human, do with it? :) That's what the harness is for: In this case, it interprets that JSON request and returns a JSON response that the model can understand.
In general, a harness is anything that manipulates the model input and output for your benefit: It may recall memories and place them into your context automatically, handle tool requests, prune long conversations, injecting parts of old conversations, and so on.
The harness is the tool you use to communicate with the model.
It can be as basic as a simple conversation tool (like ChatGPT back in the day), or more advanced like GitHub Copilot, Claude Code, etc.
Advanced harnesses will give precise instructions to the model to tell him what tooling / skills / MCPs / etc are available, and will orchestrate all the actions around the model (like starting subagents, calling external tools, etc).
The harness is the software that runs the model, and, in the case of agents, implements the agentic loop, and provides capabilities external to the model (either directly or by interfacing with other systems.)
You need a harness for ANY use of a model because otherwise you just have a bunch of numbers that are the weights and no actual software that does anything with them, and you can’t have agents without a harness that provides the capabilities that define agents.
I think that is what OP was fishing for. Wrong answers ;).
Llama.cpp is not harness, setting up prompt template to have instruct model is also not what people call harness.
What people call harness nowadays is software that you interact with and it proxies model interactions. It can do agentic loop but the loop is not really much use for models that do “thinking” unless you want to layer your own loop on top for dealing with results and good harness can help - but usually if something runs loop it is called an agent not a harness.
So if harness proxies all interaction with model it also provides tools for model. It can provide context management and security features, where as security features would be preventing injections and making sure model doesn’t access stuff it shouldn’t.
Yeah, these days if you aren't treating even the smallest of small projects that do something like "query database spit out report" as if they are major IT infrstructure projects, complete with designers, UX specialists, and accessbility experts that never talked to a disabled person to find out what they actually need, spin up 375 ec2 instances, use 19 different database systems, and send the logs to a third party, then you're literally the worst person possible, or so I've been told.
It's like, my dude last week this was an excel spreadsheet.
Given that M6 will be on TSMC smaller 2nm node and the first smaller node size in 3-years, it seems like the oddest of all years for the high-end Macs to skip.
> and now we're happy to expand the typeface with a new family called MonoLisa Text. The reasoning was to cover other use cases beyond coding with this proportional font.
Dumb question, when should a developer not use a monospaced font? I.e. when should they use MonaLisa Text
I'm sure I'm missing the obvious, but it is purely for LLM output use cases as the website implies (in which case why isn't Claude approach of using a serif font a better strategy).
Please don't take my comments are negative. Just genuinely curious, which is why I'm asking.
Since I’ve never work at FAANG, does Google have strict procedures (and approvals) before launching a product? And if so, did this go through that process?
> does Google have strict procedures (and approvals) before launching a product?
I worked at Google in the past, most recently ending in early 2015, and can confirm that the answer to this question was yes when I was there - presumably still the case today with different details.
I have no idea whether the procedures were followed in this case, nor do I have any other inside information on this story, nor am I speaking for Google or Alphabet here.
It was certainly the case for me back circa... I can barely remember, 2008/2009?
Everyone just launched tools internally, although it was pretty easy to get approval to launch something externally, although most people didn't bother. The environment back then had tons of internal tools all over the place.
I’ve been gone a few years, but there was a process for contributing OSS code outside the company, and another for releasing company code externally, etc
It seemed to mostly work. Some people complained it was too slow, others seemed to manage fine.
While I know criticizing Meta is popular, I'm not sure I'd agree with above.
Social networks didn't really exist before The Facebook. Understanding the potential market that could be created and turning down a $1B acquisition from Yahoo 20-years ago, at the time, seemed insane.
Also making the shift to mobile, when people thought that would be the death of FB is a remarkable story.
Identifying to acquire WhatsApp & Instagram, both laughed at when bought for the acquisition price at that time, now massive businesses for Meta (and their market cap value).
Meta AI glasses are surprisingly popular and growing. And more...
Note: I have no affiliation with Meta (not now or in the past)
---
EDIT: Many people I see underestimate what it takes to build a business. It is the classic “I could have built that in a weekend” critique. Maybe, but the product is only like 10% of the problem. 90% of the work (and hard part) is execution.
That's not true though. Social network sites existed, just not so "centralized" and "viral". Facebook created a simple and more user-friendly interface.
WhatsApp as it understand, does not make much if any money. Instagram is a cash cow due.
They have released some good open source technology, but as the OP said, Meta hasn't much going for it apart from addictive apps for showing ads
Social networks definitely existed, they just didn’t gain the same momentum/popularity as Facebook. As others have pointed out there was MySpace - but even before that there was Friendster and Asian Avenue. I’m sure there were probably another one or two.
That’s not to take anything away from the success of early Facebook, but the idea of a social network was not created by fb.
>Social networks didn't really exist before The Facebook.
Seriously? Of course they did. An easy counter-example is MySpace, which launched several months before Facebook's very first appearance as a Hot Or Not mimic (which predates Zuck by 3 years), and had many millions more users for years (especially while FB was restricted to colleges).
Major competition and a lot of money was going into social media around then (and for a couple years prior), FB is just the eventual winner.
Those are fair points, though I didn't say he wasn't a good business person. I could probably concede that he is, but I don't see him as the tech visionary he's often propped up to be.
> Many people I see underestimate what it takes to build a business. It is the classic “I could have built that in a weekend” critique.
My dude, no one in your reply thread is making that claim. We bristling at the claim that "Social networks didn't really exist before The Facebook". Unless you're saying FB's innovation is its business model and is what made it dominate, well, your original statement just didn't substantiate that.
IMO, the only thing I'd credit Zuck for is sticking (at least at the start) to his singular vision of what a social network should be; first it was just open for .edu emails, then when it was released more broadly their product roadmap stuck to fostering a social environment online.
And then he lost that vision. I'd say it was circa Cambridge Analytica when engagement---often ragebait because it gave them more and stronger of that sweet sweet monetizable ad signals---replaced fostering an online social environment. Others would say it's the algorithmic news feed. Either way, losing that vision started FB's demise.
IME, FB was best when it was a supplement and not a replacement for real life. FB had value to me because we could plan parties there and even keep in touch after, get that social buzz going for a little while longer. But it was _never_ the party.
But their recent efforts---Metaverse, all the AI crap---has all the hallmakrs of trying to replace real life. They now want to be the party but good luck with that. Judging by the blowback and lack of adoption consumers see what they're up to a mile away. Zuck has no idea how to stay relevant so now we have a platform more concerned with its market cap than having actual utility.
He was also a wealthy kid with connections. People place far too much weight on execution. Most of it is being in the right place at the right time and having existing connections.
I’m explicitly telling it to do something extremely specific and it’s just not listening to me.
Eg, I gave it an image to update. The image is sized 400x200 pixels. It then generates a new image at 300x300. I explicitly state to be 400x200 in size and it won’t listen.
reply