This entire situation should wake you up to how important open source models are
Get your own model running locally, go into debt if you have to
11:49 AM · Jun 13, 2026 · 14.5K Views
Forward Future's Matthew Berman and tech executive Garry Tan urge adopting local open-source AI to avoid proprietary lock-in
Tan highlighted tools like Hermes and OpenClaw for independence.
77357163624.1K
Sentiment
Positive users endorse local open source AI models as the reliable future for independence and control, while negative users dismiss them as inferior trash that lag behind frontier models and invite bans.
Pos
38.5%
Neg
61.5%
13 comments with sentiment.
Cluster Engagement
Posts from X
Most Activity
Most Activity
VIEWS189
Kapisch@onekapisch
@MatthewBerman But how can we even get closer to a model like Opus 4.8 locally?
2hViews 189Likes 4
LIKES4REPLIES4
Matthew Berman@MatthewBerman
@onekapisch It’ll happen. Just wait 6 months.
2hViews 186Likes 4
RETWEETS1
Tom Howard@_TomHoward
The cool thing about the open source harness tools like hermes, openclaw, gbrain, etc is that you totally avoid vendor lock in while still getting gains from advanced closed models.
Don’t be enslaved to any single provider, keep building up your personal AI stack.
GREG ISENBERG@gregisenberg
The takeaway from Fable 5 being BANNED by the government: GET GOOD AT LOCAL MODELS SO YOU HAVE 100% CONTROL.
My entire weekend was going to be building my craziest ideas with Fable 5. That's now cancelled.
So instead of building with Fable this weekend, I've decided I'll go deep on local models:
1. Start with the runtime. Download Ollama or LM Studio first. This is the thing that actually runs models on your machine.
2. Match the model to your hardware. A model's size is measured in billions of parameters (7B, 32B, 70B). Bigger is smarter but needs more memory. Rule of thumb: a 7B model runs on almost any laptop, a 32B needs a good Mac with 32GB+ RAM, a 70B needs serious hardware like a DGX Spark or a maxed-out Mac Studio.
3. Know which model for which job. Qwen 3 is the best all-around choice for most tasks. DeepSeek for reasoning and coding. Gemma 4 when you need something tiny that runs on a phone. Llama when you want the biggest community and the most fine-tunes.
4. Quantization. You can shrink a model to run on weaker hardware with barely any quality loss. Look for versions labeled Q4 or Q5. This is how a model that "needs" a server runs on your laptop. Learning this one concept changes everything.
5. Connect it to your agent. Point Hermes or your agent stack at a local model.
6. Context window is your real constraint locally. Cloud models give you huge context for free. Local models make you pay for it in memory. A bigger context window eats RAM fast. Keep your sessions tight and your prompts lean or your machine chokes.
7. Learn to give local models tools. A smaller local model with web search, file access, and code execution beats a giant model with none. The capability gap closes fast when you wire up the right tools. The model is the engine but the tools are the wheels.
8. Fine-tuning is more accessible than you think. You don't need this on day one, but know it exists. You can take an open model and train it on your own data so it gets good at your specific domain.
I'll probably do a breakdown at some point on this @startupideaspod if people are into it.
The lesson from this ban is basically don't build your entire workflow on something that can disappear with a single letter. Own part of your stack. Local models are insurance.
It reminds me when people realized they don't own social media accounts. And then you saw people build email lists etc.
I remember running a startup and my biggest traffic source was organic FB. All of a sudden, algo changed, and I lost 99% of my traffic.
Same sorta moment (but bigger) for AI.
This is a wake up call.
5hViews 10.6KLikes 41Bookmarks 18
VraserX e/acc@VraserX
@dedene @MatthewBerman Impossible, you’ll never have Stargate level compute at home.
1hViews 11Likes 1
Brian@BrianG12321
Quality and speed used to be the factors a year ago. Quality is almost frontier level, but speed will always be the local bottleneck cause in order to get frontier subscription level speed you need 4-8 Blackwell 6000s
1,000 tokens/second is sort of bare minimum if we're sincerely talking about replacing frontier model usage with local usage and not massively delaying productivity.
1hViews 11
Gary A.@_garytalk
@MatthewBerman Dude, I’ve been a follower from the early days. You have enough followers to not follow the herd. If I never hear “go in to debt if you have to” ever again, I’ll be happy. You have your followers, you’ve crafted a great brand. You can be yourself now 🙏
1hViews 44Likes 2
Nate Johnson@LiberN8
@MatthewBerman I joke, but it's only funny because it's kinda true.
2hViews 79Likes 1
if@ifbuilder
@MatthewBerman "Just wait 6 months" is the eternal local model promise lol. It does keep getting better though. I've run local for the privacy-sensitive stuff and it's actually usable now, just not for the heavy reasoning yet. Depends a lot on what you're building.
2hViews 64Likes 1
Pavan Reddy@pavanreddy03
@MatthewBerman What is a good hardware to buy for running a local model?
2hViews 52Likes 1
Victor Young@VictorYoungMe
@MatthewBerman Going into debt to self-host is the wrong lesson for most. It isn't "own the GPUs," it's "don't build on a model you can't swap." Running locally is one way. Designing so the model is a replaceable part is the general one, and far cheaper.
2hViews 48Likes 1
Dual Acies@DualAcies
@MatthewBerman Sell your kidney if you have to, like said by Saylor.
2hViews 125
Rick C. Hodgin@EnsignRhodium
@MatthewBerman Agree. It's the only way to be sure of what you have.
2hViews 121
Trey Taylor@TreyTaylorCEO
I keep seeing this said ... but ... If I modelmaxxed the best open source model, it's nowhere near what Fable was cranking for me. It's not like Anthropic removed all models, just the one deemed problematic and surely for a short time. I'm not even sure that the solution is to run both open and proprietary models, its just about being prepared to un up and down the model ladder.
2hViews 104
Mohammed Hossam@MohamedHz72007
@MatthewBerman I will start with a hoppy home server for 24/7 contribution to BOINC, and some hoppy back end projects, but a device that can handle at least Gemma 4 31/24 b for some kind of work is m4 mini 64 gb ram, which is not cheap, but a considerable step.
1hViews 88
John Schoenith@eleven21
@MatthewBerman I don't know, it's making me feel like it's more important to be an American right now. I know any control is not wanted or needed, but we all knew this was coming.
2hViews 59
Kristoph@kristoph
@MatthewBerman I might have heard this somewhere before @TheAhmadOsman
2hViews 41
Peter Dedene@dedene
@MatthewBerman If scaling laws are what got Anthropic and OpenAI to the current frontier, is a truly comparable local model even possible?
2hViews 40
Paul Farnam@runstop
@MatthewBerman Back frontier models outside the US. Everyone wins when the very best AI is distributed, not dominated.
1hViews 38
Harlley Oliveira | harlley.dev@harlleydev
@MatthewBerman Yes, it makes total sense for these super models to be restricted by governments. Everybody knew it would happen, they just didn't know when.
1hViews 37
Philip Reitz@Preitz00
@MatthewBerman "Go into debt if you have to." What?
2hViews 36
Load more posts