Ben

Product Manager at Tailscale

[Ben @heliostatic]

After a month, I feel more comfortable searching with @Neeva than I have in years of using Google, DuckDuckGo, Cuil (RIP), and others. Paying for a product makes the relationship clear. Everything old is new. My referral link: https://t.co/9oyiRzWmxK or try it without my code :)

[Ben @heliostatic]

@molly0xFFF Oooh! Congratulations!

[Ben @heliostatic]

@JillWohlner @thegoodrc Congratulations! Bookmarking for all future startup hiring needs.

[Tailscale @Tailscale]

Launch day: we're introducing grants, the next generation of ACLs. With grants, you can manage access controls from the network layer into the application layer. https://t.co/s5LK2SpzLC

[Tailscale @Tailscale]

New: Restrict access for non-compliant devices with Tailscale Device Posture Management, now available in beta! https://t.co/fDz4003A6u

[Ben @heliostatic]

@DanielMiessler Kagi has been my replacement. Haven't had to switch to Google except for the rare case I'm actually doing shopping in a new category and want ads (shout-out Google Shopping team)

[Dan Siroker @dsiroker]

Wow! Over 3M views on social media and 3,177 preorders and growing! 📈 Thank you!!!

Here's the answer to the question: how can we prevent people from being recorded without their consent?

These are the kind of features the device in that Black Mirror episode should have had!

[eigenrobot @eigenrobot]

occasional reminder that lord of the flies was written because Golding read about boys being marooned and having a good time and this made him angry because he hated children so he wrote a book about dudes being assholes

but no. irl, marooned bros on islands are chill

[Itamar Golan 🤓 @ItakGol]

Prompt-Visual Injection is the new kid in town! 😈🛡

[Dan Siroker @dsiroker]

🥳Big news: @RewindAI is coming to Windows!

Today I shared the keynote stage with Intel CEO @PGelsinger to demo Rewind on Windows.

The demo includes Ask Rewind powered by Llama 2 running entirely locally!

Bonus: As of today, Rewind can also run on Intel-based Macs!

(1/4)

[Ben @heliostatic]

@CubicleApril @__apf__ Only 6! Such alacrity! Our town finished one this year that was initially proposed in 1912. Good old town meeting direct democracy.

[Ed Zitron @edzitron]

Sir that is bullshit, according to my source “you”

https://t.co/VcWPVxubG7

[[email protected] @willnorris]

@stkenned @Tailscale @Cloudflare you mean https://t.co/oT1K8DIlYL ?

[Max Flugrath🗳️ @MaxFlugrath]

New Veep deleted scene. 🍧

[Dan Siroker @dsiroker]

Introducing Rewind for iPhone - a truly personalized AI in your pocket!

🔍 Browse & search for anything you’ve seen (including screenshots)
🤖 Summarize and ask any question using AI
🔒 Private by design

Learn more: https://t.co/05G3jkVwqw

(1/6)

[Kaylan_TX @Kaylan_TX_]

@TwitterBusiness “Twitter is healthy”

Ya, I wouldn’t wanna rush to judgement calling this man a Nazi just because of swastikas in his messages or because of the death threats he sent. It’s clearly my lack of intelligence & deep thinking on his POV that are the problem here🙃

[Yam Peleg @Yampeleg]

I went through many open-source models lately.

Here are my current top models that I suggest you test for yourself:

- Nous-Hermes: Still the best in my opinion for day-to-day usecases. It follows your instructions flawlessly nearly all the time. [Especially if you use beam search]

Link: https://t.co/bmj0Xp0oIm

- WizardCoder: Yes. It is a coding model but wow it is also so damn powerful on everything. I was floored by it when I first tried it.

Link: https://t.co/gOAo7R3h5p

- OpenChat V2 Weighted: Leading AlpacaEval. Really cool model and the approach behind it (Offline RL) is currently something I try myself based on their write up.

Link: https://t.co/XSq74g0LkO

- Baichuan-13B-Chat: A very good model! Following some of the most tricky instructions I tried with ease.

Link: https://t.co/NvvsdiS0WZ

- ChatGLM2-6B: Amazing model that is hard to believe is this small. It's training method is also very interesting! (incorporating MLM pretraining with a causal model)

Link: https://t.co/xsKgDcGQTE

- UltraLM-13b: The data this model trained on is the best I had seen on Huggingface. Hands down. The model itself got some criticism lately but I think that it should get a second chance.

Link: https://t.co/pmMlLn2i7t

- InternLM-7b: Metric-wise: This model is the best 7B model we got today. It is also very good on day to day use cases but just take in consideration that you need to allow external source code to run if you want to use it.

Link: https://t.co/98O2ftwWlg

Summary:

I deliberately did not include models that are trained "LIMA style" (Small amount of data).
They might be good but I think they miss the point: This is data science. It is all about the data.

We should strive to use as much data as possible that contain as much information and facts as possible.

Coming up with unique ways to use less data is absolutely good but I think this should be done as the last step of the training.

To "align" your model to answer nicely, so I am more interested in small models that are trained on huge datasets first.

Why small models?

Most models at 65B are much better mostly because of their scale. (Insane model to try: Airoboros 65B)

Opinion about scale:
We should absolutely aim to train a huge SOTA model open source once we clear all the details training smaller models first!

In the picture: OpenChat V2-W overtaking ChatGPT-3.5 on AlpacaEval. I don't think it means much but I like looking at this! 🙃

[apenwarr @apenwarr]

This was a great talk from @theprincessxena about using the tsnet library to put tailscale directly in your app, thus avoiding annoyances like firewalls, extra daemons in your containers, etc. https://t.co/jlxgR7SUdR

[Alex Widua @alexwidua]

Prototyped a little app that allows you to take frames from @figma and 'pull them into space'.

The frames stay linked to the canvas and update on any changes

[Nathalie Baptiste 🇭🇹 @nhbaptiste]

Ah, well