web dev, digital marketer, content creator, graphic designer, solo-founder | dad/husband | prev. nightclub bartender, property manager
NW FL
- Tweets 64,374
- Following 1,360
- Followers 2,439
- Likes 107,735
d3layd retweeted
hi haiku 🥹 such a great model for tasks that are mostly reading/small edits/tool calls and don't need much reasoning!
Introducing Claude Haiku 5.5: the cheapest, fastest, and most capable small model we’ve ever released.
On average, it costs around 75% less to run than Claude Haiku 4.5.
74
20
794
23,966
IDK who you are, but if @TheWarpedJoe doesn't like you, you must really suck.
I can be an excessively petty person when the target has caused harm to people I consider to be a friend.
I do not call many people "friend".
This is targeted to one person and I know this person reads my tweets
1/
4
45
d3layd retweeted
For prompts under 100K tokens, it's $0.10 / $0.50 per million tokens with cache reads at $0.01.
Above 100K tokens, $0.50 / $2.50 per million, with cache reads at $0.05.
15
14
828
172,605
d3layd retweeted
Haiku 5.5 is now available in the Claude Platform and Claude Code. On average, it costs around 75% less to run than Haiku 4.5.
It pairs well with Opus 5.5 or Sonnet 5.5 as a subagent. Use it for high-volume, cost-sensitive tasks like summaries, compactions, or database queries.
Introducing Claude Haiku 5.5: the cheapest, fastest, and most capable small model we’ve ever released.
On average, it costs around 75% less to run than Claude Haiku 4.5.
207
467
8,400
496,890
d3layd retweeted
all those years of refusing to learn math about to payoff
We’re releasing a broad range of new mathematical results produced by an internal frontier model.
We’ve been consulting with the independent Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study, and we have drawn on their advice and public recommendations to inform how we release these results.
github.com/openai/math
17
9
365
9,294
We improve performance by orders of magnitude here, SIR
boat.dev's VMs now have 11.5× faster first agent answer on resume/fork
sandbox latency isn't about boot time at all
when you boot from a snapshot, the VM might boot fast
but running stuff on it initially might still be slow as the files are lazily restored in the background
so we've been making that process way more optimized over time
median time, before → now:
- Claude Code, first answer: 102s → 8.8s (−91%)
- pi, first answer: 127s → 13s (−90%)
- Codex, first answer: 62s → 8.1s (−87%)
- Android, emulator ready: 115s → 47s (−60%)
- Web apps, first page: 16s → 14s (−16%)
p90s also got shrunk to quite near the median:
- Claude Code: 120s → 12s (−90%)
- pi: 127s → 60s (−53%)
- Codex: 114s → 9.3s (−92%)
- Android: 155s → 52s (−66%)
- Web apps: 24s → 19s (−23%)
this is still to be improved further with caching! these are fully uncached times, requiring full snapshot transfer over the wire from object storage
1
1
116
Scary, but also I wonder if AI will actually allow us to identify these backdoors, how they made it into the supply chain, and harden the chain itself moving forward.
Or, will it just get exponentially worse?
BREAKING: Dutch intelligence says Chinese hackers hold secret source code of proprietary Western hardware and software they target, letting them dig deep into it for weaknesses.
AIVD and MIVD expect attacks on firewalls and VPN gateways to rise sharply in the coming years, with AI speeding up exploitation.
25
d3layd retweeted
As reported in the @WSJ, we have raised a Series B to bring Hermes Agent to new frontiers (and, yes, build a mobile app)
Thank you to our investors including @nvidia @M12vc @SamsungNext @robotventures @usv @ycombinator @MenloVentures, among others
wsj.com/pro/venture-capital/…
sharing my thoughts on our recent fundraise.
we did alright for a couple of goofballs
nousresearch.com/a-note-on-o…
245
193
2,580
101,628
This, but realize that what works for one business, won't necessarily work for another.
For certain industries, marketing is better spent on networking and going to events. For others it's referrals and affiliates, for others it's just content.
One thing I know for a fact, don't start experimenting with a new marketing trend because you have FOMO. Don't ever do that. Find what works for you and grow naturally.
Hot take:
you’re not building a real product if you’re not marketing it.
you’re just working on your hobby.
20
d3layd retweeted
Separating red team and defense access imo shows a very poor understanding of what it means to work in security.
How can you “defend” your product if you can’t attempt offensive actions?
We’re expanding our Cyber Verification Program to give security professionals broader access to our most capable models.
Through this program, verified security professionals can access Claude Mythos 5.1, Opus 5.5, and Sonnet 5.5 with safeguards designed for defensive work.
We’re also opening up new tiers to allow for authorized offensive work, like penetration testing and red-teaming.
anthropic.com/news/cyber-ver…
33
43
425
11,136
tfw you pull up to the trap and end up snagging some sick kicks
Yall said he wouldn’t deliver on product, but look what came in the mail @based_d0m niggabutt retro 3s
1
49